Merge origin/master after #784 ; bound test 102's crawls

Every other network crawl runs under a deadline through local-crawl.sh. This test drives its own FTP fixture directly, so a wedge would hang the job until CI cancels it and discards the log (#796). Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Xavier Roche <roche@httrack.com>
hts_rename_over() deletes its destination when the source is missing (#784 )
2026-07-27 19:12:54 +03:00 · 2026-07-27 12:17:46 +02:00 · 2026-07-27 12:07:12 +02:00 · 2026-07-27 11:44:43 +02:00 · 2026-07-27 11:31:47 +02:00 · 2026-07-27 11:15:13 +02:00
38 changed files with 1409 additions and 353 deletions
--- a/.github/workflows/windows-build.yml
+++ b/.github/workflows/windows-build.yml
@@ -225,11 +225,12 @@ jobs:

          # Every gate here exits 77, so an all-skipped suite would report green having
          # tested nothing: pin the skips, and floor the passes in case the glob empties.
-          # footer-overflow skips on Windows (needs a path past MAX_PATH); crange pending #581;
+          # footer-overflow and purge-longpath skip on Windows (need a path past MAX_PATH);
+          # crange pending #581;
          # webdav-mime needs a reapable background listener, which MSYS cannot give it;
          # badmtime needs a filesystem that stores an mtime past gmtime's range;
          # single-file ends on a GUI half needing htsserver, which this job does not build.
-          expected_skips=" 01_engine-footer-overflow.test 48_local-crange-memresume.test 71_local-crange-repaircache.test 79_local-proxytrack-webdav-mime.test 88_local-proxytrack-badmtime.test 94_local-single-file.test"
+          expected_skips=" 01_engine-footer-overflow.test 100_local-purge-longpath.test 48_local-crange-memresume.test 71_local-crange-repaircache.test 79_local-proxytrack-webdav-mime.test 88_local-proxytrack-badmtime.test 94_local-single-file.test"
          [ "$pass" -ge 90 ] || { echo "::error::only $pass tests passed ($skip skipped)"; exit 1; }
          [ "$skipped" = "$expected_skips" ] || { echo "::error::unexpected skips:$skipped"; exit 1; }
          [ "$fail" -eq 0 ] || { echo "::error::failing:$failed"; exit 1; }
--- a/src/hts-indextmpl.h
+++ b/src/hts-indextmpl.h
@@ -487,141 +487,6 @@ regen:

 #define HTS_DATA_UNKNOWN_HTML_LEN 0

-#define HTS_DATA_ERROR_HTML "<html>"LF\
-  "<html xmlns=\"http://www.w3.org/1999/xhtml\" lang=\"en\">"LF\
-  ""LF\
-  "<head>"LF\
-  "	<meta http-equiv=\"Content-Type\" content=\"text/html; charset=utf-8\" />"LF\
-  "	<meta name=\"description\" content=\"HTTrack is an easy-to-use website mirror utility. It allows you to download a World Wide website from the Internet to a local directory,building recursively all structures, getting html, images, and other files from the server to your computer. Links are rebuiltrelatively so that you can freely browse to the local site (works with any browser). You can mirror several sites together so that you can jump from one toanother. You can, also, update an existing mirror site, or resume an interrupted download. The robot is fully configurable, with an integrated help\" />"LF\
-  "	<meta name=\"keywords\" content=\"httrack, HTTRACK, HTTrack, winhttrack, WINHTTRACK, WinHTTrack, offline browser, web mirror utility, aspirateur web, surf offline, web capture, www mirror utility, browse offline, local  site builder, website mirroring, aspirateur www, internet grabber, capture de site web, internet tool, hors connexion, unix, dos, windows 95, windows 98, solaris, ibm580, AIX 4.0, HTS, HTGet, web aspirator, web aspirateur, libre, GPL, GNU, free software\" />"LF\
-  "	<title>Page not retrieved! - HTTrack Website Copier</title>"LF\
-  "	<style type=\"text/css\">"LF\
-  "	<!--"LF\
-  ""LF\
-  "body {"LF\
-  "	margin: 0;  padding: 0;  margin-bottom: 15px;  margin-top: 8px;"LF\
-  "	background: #77b;"LF\
-  "}"LF\
-  "body, td {"LF\
-  "	font: 14px \"Trebuchet MS\", Verdana, Arial, Helvetica, sans-serif;"LF\
-  "	}"LF\
-  ""LF\
-  "#subTitle {"LF\
-  "	background: #000;  color: #fff;  padding: 4px;  font-weight: bold; "LF\
-  "	}"LF\
-  ""LF\
-  "#siteNavigation a, #siteNavigation .current {"LF\
-  "	font-weight: bold;  color: #448;"LF\
-  "	}"LF\
-  "#siteNavigation a:link    { text-decoration: none; }"LF\
-  "#siteNavigation a:visited { text-decoration: none; }"LF\
-  ""LF\
-  "#siteNavigation .current { background-color: #ccd; }"LF\
-  ""LF\
-  "#siteNavigation a:hover   { text-decoration: none;  background-color: #fff;  color: #000; }"LF\
-  "#siteNavigation a:active  { text-decoration: none;  background-color: #ccc; }"LF\
-  ""LF\
-  ""LF\
-  "a:link    { text-decoration: underline;  color: #00f; }"LF\
-  "a:visited { text-decoration: underline;  color: #000; }"LF\
-  "a:hover   { text-decoration: underline;  color: #c00; }"LF\
-  "a:active  { text-decoration: underline; }"LF\
-  ""LF\
-  "#pageContent {"LF\
-  "	clear: both;"LF\
-  "	border-bottom: 6px solid #000;"LF\
-  "	padding: 10px;  padding-top: 20px;"LF\
-  "	line-height: 1.65em;"LF\
-	" background-image: url(backblue.gif);"LF\
-	" background-repeat: no-repeat;"LF\
-	" background-position: top right;"LF\
-  "	}"LF\
-  ""LF\
-  "#pageContent, #siteNavigation {"LF\
-  "	background-color: #ccd;"LF\
-  "	}"LF\
-  ""LF\
-  ""LF\
-  ".imgLeft  { float: left;   margin-right: 10px;  margin-bottom: 10px; }"LF\
-  ".imgRight { float: right;  margin-left: 10px;   margin-bottom: 10px; }"LF\
-  ""LF\
-  "hr { height: 1px;  color: #000;  background-color: #000;  margin-bottom: 15px; }"LF\
-  ""LF\
-  "h1 { margin: 0;  font-weight: bold;  font-size: 2em; }"LF\
-  "h2 { margin: 0;  font-weight: bold;  font-size: 1.6em; }"LF\
-  "h3 { margin: 0;  font-weight: bold;  font-size: 1.3em; }"LF\
-  "h4 { margin: 0;  font-weight: bold;  font-size: 1.18em; }"LF\
-  ""LF\
-  ".blak { background-color: #000; }"LF\
-  ".hide { display: none; }"LF\
-  ".tableWidth { min-width: 400px; }"LF\
-  ""LF\
-  ".tblRegular       { border-collapse: collapse; }"LF\
-  ".tblRegular td    { padding: 6px;  background-image: url(fade.gif);  border: 2px solid #99c; }"LF\
-  ".tblHeaderColor, .tblHeaderColor td { background: #99c; }"LF\
-  ".tblNoBorder td   { border: 0; }"LF\
-  ""LF\
-  ""LF\
-  "// -->"LF\
-  "</style>"LF\
-  ""LF\
-  "</head>"LF\
-  ""LF\
-  "<table width=\"76%%\" border=\"0\" align=\"center\" cellspacing=\"0\" cellpadding=\"3\" class=\"tableWidth\">"LF\
-  "	<tr>"LF\
-  "	<td id=\"subTitle\">HTTrack Website Copier - Open Source offline browser</td>"LF\
-  "	</tr>"LF\
-  "</table>"LF\
-  "<table width=\"76%%\" border=\"0\" align=\"center\" cellspacing=\"0\" cellpadding=\"0\" class=\"tableWidth\">"LF\
-  "<tr class=\"blak\">"LF\
-  "<td>"LF\
-  "	<table width=\"100%%\" border=\"0\" align=\"center\" cellspacing=\"1\" cellpadding=\"0\">"LF\
-  "	<tr>"LF\
-  "	<td colspan=\"6\"> "LF\
-  "		<table width=\"100%%\" border=\"0\" align=\"center\" cellspacing=\"0\" cellpadding=\"10\">"LF\
-  "		<tr> "LF\
-  "		<td id=\"pageContent\"> "LF\
-  "<!-- ==================== End prologue ==================== -->"LF\
-  "<h1><strong><u>Oops!...</u></strong></h1>"LF\
-  "<h3>This page has <font color=\"red\"><em>not</em></font> been retrieved by HTTrack Website Copier (%s). </h3>"LF\
-  "<script language=\"Javascript\">"LF\
-  "<!--"LF\
-  "  var loc=document.location.toString();"LF\
-  "  if (loc) {"LF\
-  "    var pos=loc.indexOf('link=');"LF\
-  "    if (pos>0) {"LF\
-  "      document.write('Clic to the link <b>below</b> to go to the online location!<br><a href=\"'+loc.substring(pos+5)+'\">'+loc.substring(pos+5)+'</a><br>');"LF\
-  "    } else"LF\
-  "      document.write('(no location defined)');"LF\
-  "  }"LF\
-  "// -->"LF\
-  "</script>"LF\
-  "<h6 align=\"right\">Mirror by HTTrack Website Copier</h6>"LF\
-  "</body>"LF\
-  "</html>"LF\
-  "<!-- ==================== Start epilogue ==================== -->"LF\
-  "		</td>"LF\
-  "		</tr>"LF\
-  "		</table>"LF\
-  "	</td>"LF\
-  "	</tr>"LF\
-  "	</table>"LF\
-  "</td>"LF\
-  "</tr>"LF\
-  "</table>"LF\
-  ""LF\
-  "<table width=\"76%%\" height=\"100%%\" border=\"0\" align=\"center\" valign=\"bottom\" cellspacing=\"0\" cellpadding=\"0\">"LF\
-  "	<tr>"LF\
-  "	<td id=\"footer\"><small>&copy; 2014 Xavier Roche & other contributors - Web Design: Kauler Leto.</small></td>"LF\
-  "	</tr>"LF\
-  "</table>"LF\
-  ""LF\
-  "</body>"LF\
-  ""LF\
-  "</html>"LF\
-  ""LF\
-  ""LF
-
 // image gif "unknown"
 #define HTS_DATA_UNKNOWN_GIF \
  "\x47\x49\x46\x38\x39\x61\x20\x0\x20\x0\xf7\xff\x0\xc0\xc0\xc0\xff\x0\x0\xfc\x3\x0\xf8\x6\x0\xf6\x9\x0\xf2\xc\x0\xf0\xf\x0\xf0\xe\x0\xed\x11\x0\xec\x13\x0\xeb\x14\x0\xe9\x15\x0\xe8\x18\x0\xe6\x18\x0\xe5\x1a\x0\xe3\x1c\x0\xe2\x1d\x0\xe1\x1e\x0\xdf\x20\x0\xdd\x23\x0\xdd\x22\x0\xdb\x23\x0\xda\x25\x0\xd9\x25\x0\xd8\x27\x0\xd6\x29\x0\xd5\x2a\x0\xd3\x2c\x0\xd2\x2d\x0"\
--- a/src/htsback.c
+++ b/src/htsback.c
@@ -340,12 +340,61 @@ static int back_index_ready(httrackp * opt, struct_back * sback, const char *adr
 }

 static int slot_can_be_cached_on_disk(const lien_back * back) {
+  /* A pending backup or spool means the slot is not finalized, and the swap
+     would unlink it through back_clear_entry() (#771). */
+  if (back->tmpfile != NULL && back->tmpfile[0] != '\0')
+    return 0;
  return (back->status == STATUS_READY && back->locked == 0
          && back->url_sav[0] != '\0'
          && strcmp(back->url_sav, BACK_ADD_TEST) != 0);
  /* Note: not checking !IS_DELAYED_EXT(back->url_sav) or it will quickly cause the slots to be filled! */
 }

+int back_selftest_slot_swap(void) {
+  lien_back back;
+  int err = 0;
+
+#define CHECK(want, why)                                                       \
+  do {                                                                         \
+    if (slot_can_be_cached_on_disk(&back) != (want)) {                         \
+      fprintf(stderr, "backswap: expected %d for %s\n", (want), (why));        \
+      err = 1;                                                                 \
+    }                                                                          \
+  } while (0)
+
+  memset(&back, 0, sizeof(back));
+  back.status = STATUS_READY;
+  strcpybuff(back.url_sav, "/tmp/httrack-selftest.bin");
+  CHECK(1, "a plain ready slot");
+
+  back.tmpfile = back.tmpfile_buffer;
+  strcpybuff(back.tmpfile_buffer, "/tmp/httrack-selftest.bin.bak");
+  CHECK(0, "a slot still holding a re-fetch backup");
+
+  /* Callers clear a spent temporary by emptying the name, not the pointer. */
+  back.tmpfile_buffer[0] = '\0';
+  CHECK(1, "a slot whose temporary was already dropped");
+
+  back.tmpfile = NULL;
+  back.locked = 1;
+  CHECK(0, "a locked slot");
+  back.locked = 0;
+
+  back.status = STATUS_TRANSFER;
+  CHECK(0, "a slot still transferring");
+  back.status = STATUS_READY;
+
+  back.url_sav[0] = '\0';
+  CHECK(0, "a slot with no save name");
+
+  strcpybuff(back.url_sav, BACK_ADD_TEST);
+  CHECK(0, "the dummy test slot");
+#undef CHECK
+
+  printf("backswap self-test: %s\n", err ? "FAIL" : "OK");
+  return err;
+}
+
 /* Put all backing entries that are ready in the storage hashtable to spare space and CPU */
 int back_cleanup_background(httrackp * opt, cache_back * cache,
                            struct_back * sback) {
@@ -585,6 +634,42 @@ static int create_back_tmpfile(httrackp *opt, lien_back *const back,
  return 0;
 }

+/* Note: utf-8 */
+void back_refetch_backup(httrackp *opt, lien_back *const back) {
+  back->tmpfile = NULL;
+  if (fexist_utf8(back->url_sav)) {
+    hts_boolean saved = HTS_FALSE;
+
+    if (create_back_tmpfile(opt, back, "bak") == 0) {
+      /* clobber a .bak a killed run left behind, or the guard stays off for
+         good (#758) */
+      if (fexist_utf8(back->tmpfile))
+        hts_log_print(opt, LOG_WARNING, "replacing leftover backup %s",
+                      back->tmpfile);
+      saved = hts_rename_over(back->url_sav, back->tmpfile);
+    }
+    if (!saved) {
+      hts_log_print(opt, LOG_WARNING | LOG_ERRNO,
+                    "could not back up %s; an aborted re-fetch will lose it",
+                    back->url_sav);
+      back->tmpfile = NULL;
+    }
+  }
+}
+
+/* Did the fetch fail to produce a response, as opposed to the engine
+   deliberately passing the resource over? Only the latter may be purged. */
+static hts_boolean back_transfer_failed(const int statuscode) {
+  switch (statuscode) {
+  case STATUSCODE_TOO_BIG:
+  case STATUSCODE_EXCLUDED:
+  case STATUSCODE_TEST_OK:
+    return HTS_FALSE;
+  default:
+    return statuscode <= 0 ? HTS_TRUE : HTS_FALSE;
+  }
+}
+
 /* Commit or restore a re-fetch backup (#77 follow-up): a re-fetch over an
   existing file moved the good copy to back->tmpfile before truncating url_sav.
   commit keeps the new file and drops the backup; else restore it so an aborted
@@ -1031,6 +1116,14 @@ int back_finalize(httrackp * opt, cache_back * cache, struct_back * sback,
  /* Aborted, error, or not ready: url_sav (if written) is broken; restore the
     previous copy from the backup. */
  back_finalize_backup(opt, &back[p], HTS_FALSE);
+  /* Note the surviving copy, or the end-of-update purge drops what this run
+     never managed to replace (#746). */
+  if (!back[p].testmode && back_transfer_failed(back[p].r.statuscode) &&
+      back[p].url_sav[0] != '\0' && fexist_utf8(back[p].url_sav)) {
+    filenote(&opt->state.strc, back[p].url_sav, NULL);
+    file_notify(opt, back[p].url_adr, back[p].url_fil, back[p].url_sav, 0, 0,
+                back[p].r.notmodified);
+  }
  return -1;
 }

@@ -3143,20 +3236,7 @@ void back_wait(struct_back * sback, httrackp * opt, cache_back * cache,
                                          back[i].url_sav, 1, 1,
                                          back[i].r.notmodified);
                              back[i].r.compressed = 0;
-                              /* Re-fetch over an existing file (#77 follow-up):
-                                 move the good copy aside before truncating it
-                                 so an aborted transfer can restore it. url_sav
-                                 is still written normally (file list intact).
-                               */
-                              back[i].tmpfile = NULL;
-                              if (fexist_utf8(back[i].url_sav)) {
-                                if (create_back_tmpfile(opt, &back[i], "bak") !=
-                                        0 ||
-                                    RENAME(back[i].url_sav, back[i].tmpfile) !=
-                                        0) {
-                                  back[i].tmpfile = NULL;
-                                }
-                              }
+                              back_refetch_backup(opt, &back[i]);
                              if ((back[i].r.out =
                                   filecreate(&opt->state.strc,
                                              back[i].url_sav)) == NULL) {
--- a/src/htsback.h
+++ b/src/htsback.h
@@ -139,6 +139,12 @@ int back_trylive(httrackp * opt, cache_back * cache, struct_back * sback,
                 const int p);
 int back_finalize(httrackp * opt, cache_back * cache, struct_back * sback,
                  const int p);
+/* Move the previous copy of back->url_sav to back->tmpfile so back_finalize()
+   can put it back when the re-fetch fails (#77 follow-up). Call right before
+   truncating url_sav; tmpfile stays NULL when there is nothing to save. */
+void back_refetch_backup(httrackp *opt, lien_back *const back);
+/* -#test=backswap: slots eligible for the on-disk ready table. */
+int back_selftest_slot_swap(void);
 void back_info(struct_back * sback, int i, int j, FILE * fp);
 void back_infostr(struct_back *sback, int i, int j, char *s, size_t size);
 LLint back_transferred(LLint add, struct_back * sback);
--- a/src/htscache.c
+++ b/src/htscache.c
@@ -855,13 +855,6 @@ static htsblk cache_readex_new(httrackp * opt, cache_back * cache,
  return r;
 }

-// lecture d'un fichier dans le cache
-// si save==null alors test unqiquement
-static int hts_rename(httrackp * opt, const char *a, const char *b) {
-  hts_log_print(opt, LOG_DEBUG, "Cache: rename %s -> %s (%p %p)", a, b, a, b);
-  return RENAME(a, b);
-}
-
 /* Open the cache ZIP via hts_fopen_utf8 so a non-ASCII path_log isn't mangled
   to ANSI (#630); 64-bit funcs keep multi-GB caches whole on Windows LLP64. */
 static voidpf ZCALLBACK hts_zip_fopen_utf8(voidpf opaque, const void *filename,
@@ -991,29 +984,15 @@ void cache_init(cache_back * cache, httrackp * opt) {
              OPT_GET_BUFF(opt), OPT_GET_BUFF_SIZE(opt),
              StringBuff(opt->path_log),
              "hts-cache/new.zip")))) { // a previous cache exists.. rename it
-        /* Remove OLD cache */
-        if (fexist_utf8(fconcat(OPT_GET_BUFF(opt), OPT_GET_BUFF_SIZE(opt),
-                                StringBuff(opt->path_log),
-                                "hts-cache/old.zip"))) {
-          if (UNLINK(fconcat(OPT_GET_BUFF(opt), OPT_GET_BUFF_SIZE(opt),
-                             StringBuff(opt->path_log), "hts-cache/old.zip")) !=
-              0) {
-            hts_log_print(opt, LOG_WARNING | LOG_ERRNO,
-                          "Cache: error while moving previous cache");
-          }
-        }
-
-        /* Rename */
-        if (hts_rename
-            (opt,
-             fconcat(OPT_GET_BUFF(opt), OPT_GET_BUFF_SIZE(opt), StringBuff(opt->path_log),
-                     "hts-cache/new.zip"), fconcat(OPT_GET_BUFF(opt), OPT_GET_BUFF_SIZE(opt),
-                                                   StringBuff(opt->path_log),
-                                                   "hts-cache/old.zip")) != 0) {
+        if (!hts_rename_over(
+                fconcat(OPT_GET_BUFF(opt), OPT_GET_BUFF_SIZE(opt),
+                        StringBuff(opt->path_log), "hts-cache/new.zip"),
+                fconcat(OPT_GET_BUFF(opt), OPT_GET_BUFF_SIZE(opt),
+                        StringBuff(opt->path_log), "hts-cache/old.zip"))) {
          hts_log_print(opt, LOG_WARNING | LOG_ERRNO,
                        "Cache: error while moving previous cache");
        } else {
-          hts_log_print(opt, LOG_DEBUG, "Cache: successfully renamed");
+          hts_log_print(opt, LOG_DEBUG, "Cache: rotated new.zip to old.zip");
        }
      }
    } else {
--- a/src/htscache_selftest.c
+++ b/src/htscache_selftest.c
@@ -676,7 +676,7 @@ int cache_selftests(httrackp *opt, const char *dir) {
    char base[HTS_URLMAXSIZE];

    strcpybuff(base, dir);
-    if (base[0] != '\0' && base[strlen(base) - 1] != '/') {
+    if (base[0] != '\0' && hts_lastchar(base) != '/') {
      strcatbuff(base, "/");
    }
    StringCopy(opt->path_log, base);
@@ -856,7 +856,7 @@ static void golden_setup(httrackp *opt, const char *dir) {
  char base[HTS_URLMAXSIZE];

  strcpybuff(base, dir);
-  if (base[0] != '\0' && base[strlen(base) - 1] != '/') {
+  if (base[0] != '\0' && hts_lastchar(base) != '/') {
    strcatbuff(base, "/");
  }
  StringCopy(opt->path_log, base);
--- a/src/htscore.c
+++ b/src/htscore.c
@@ -773,7 +773,7 @@ int httpmirror(char *url1, httrackp * opt) {
        // sauter les + sans rien après..
        if (strnotempty(tempo)) {
          if ((plus == 0) && (type == 1)) {     // implicite: *www.edf.fr par exemple
-            if (tempo[strlen(tempo) - 1] != '*') {
+            if (hts_lastchar(tempo) != '*') {
              strcatbuff(tempo, "*");   // ajouter un *
            }
          }
@@ -1819,60 +1819,6 @@ int httpmirror(char *url1, httrackp * opt) {
        // -- -- --
        // sauver fichier

-        /* En cas d'erreur, vérifier que fichier d'erreur existe */
-        if (strnotempty(savename()) == 0) {       // chemin de sauvegarde existant
-          if (strcmp(urlfil(), "/robots.txt") == 0) {     // pas robots.txt
-            if (store_errpage) {        // c'est une page d'erreur
-              int create_html_warning = 0;
-              int create_gif_warning = 0;
-
-              switch (ishtml(opt, urlfil())) {    /* pas fichier html */
-              case 0:          /* non html */
-                {
-                  char buff[256];
-
-                  guess_httptype_sized(opt, buff, sizeof(buff), urlfil());
-                  if (strcmp(buff, "image/gif") == 0)
-                    create_gif_warning = 1;
-                }
-                break;
-              case 1:          /* html */
-                if (!r.adr) {
-                }
-                break;
-              default:         /* don't know.. */
-                break;
-              }
-              /* Créer message d'erreur ? */
-              if (create_html_warning) {
-                char *adr =
-                  (char *) malloct(strlen(HTS_DATA_ERROR_HTML) + 1100);
-                hts_log_print(opt, LOG_INFO, "Creating HTML warning file (%s)",
-                              r.msg);
-                if (adr) {
-                  if (r.adr) {
-                    freet(r.adr);
-                    r.adr = NULL;
-                  }
-                  sprintf(adr, HTS_DATA_ERROR_HTML, r.msg);
-                  r.adr = adr;
-                }
-              } else if (create_gif_warning) {
-                char *adr = (char *) malloct(HTS_DATA_UNKNOWN_GIF_LEN);
-
-                hts_log_print(opt, LOG_INFO, "Creating GIF dummy file (%s)",
-                              r.msg);
-                if (r.adr) {
-                  freet(r.adr);
-                  r.adr = NULL;
-                }
-                memcpy(adr, HTS_DATA_UNKNOWN_GIF, HTS_DATA_UNKNOWN_GIF_LEN);
-                r.adr = adr;
-              }
-            }
-          }
-        }
-
        if (strnotempty(savename()) == 0) {       // pas de chemin de sauvegarde
          if (strcmp(urlfil(), "/robots.txt") == 0) {     // robots.txt
            char BIGSTK sitemaps[8192];
@@ -2190,7 +2136,9 @@ int httpmirror(char *url1, httrackp * opt) {
                  continue;
                strcpybuff(file, StringBuff(opt->path_html));
                strcatbuff(file, line + 1);
-                file[strlen(file) - 1] = '\0';
+                /* strip filenote()'s ']', absent when linput() truncated the
+                   line */
+                hts_striplastchar(file, ']');
                hts_changes_previous(opt, file + StringLength(opt->path_html));
                if (!strstr(adr, line)) { // not found in the new list?
                  if (fexist_utf8(file)) { // still on disk
@@ -2217,12 +2165,11 @@ int httpmirror(char *url1, httrackp * opt) {
                fseek(old_lst, 0, SEEK_SET);
                while(!feof(old_lst)) {
                  linput(old_lst, line, 1000);
-                  while(strnotempty(line) && (line[strlen(line) - 1] != '/')
-                        && (line[strlen(line) - 1] != '\\')) {
-                    line[strlen(line) - 1] = '\0';
+                  while (strnotempty(line) && (hts_lastchar(line) != '/') &&
+                         (hts_lastchar(line) != '\\')) {
+                    hts_choplastchar(line);
                  }
-                  if (strnotempty(line))
-                    line[strlen(line) - 1] = '\0';
+                  hts_choplastchar(line);
                  if (strnotempty(line))
                    if (!strstr(adr, line)) {   // non trouvé?
                      char BIGSTK file[HTS_URLMAXSIZE * 2];
@@ -2235,13 +2182,12 @@ int httpmirror(char *url1, httrackp * opt) {
                        if (opt->log) {
                          hts_log_print(opt, LOG_INFO, "Purging directory %s/",
                                        file);
-                          while(strnotempty(file)
-                                && (file[strlen(file) - 1] != '/')
-                                && (file[strlen(file) - 1] != '\\')) {
-                            file[strlen(file) - 1] = '\0';
+                          while (strnotempty(file) &&
+                                 (hts_lastchar(file) != '/') &&
+                                 (hts_lastchar(file) != '\\')) {
+                            hts_choplastchar(file);
                          }
-                          if (strnotempty(file))
-                            file[strlen(file) - 1] = '\0';
+                          hts_choplastchar(file);
                        }
                      }
                    }
--- a/src/htscore.h
+++ b/src/htscore.h
@@ -376,8 +376,8 @@ void hts_finish_makeindex(httrackp *opt, int *makeindex_done,
                          const char *template_footer, const char *adr,
                          const char *fil);

-// Flush ht_buff[0..ht_len] to save on disk (skip if MD5 unchanged); *fp
-// closed+NULLed on write. Precondition: ht_len>0.
+// Flush ht_buff[0..ht_len] to save on disk; *fp closed+NULLed on write.
+// Precondition: ht_len>0.
 void hts_finish_html_file(httrackp *opt, cache_back *cache, htsblk *r,
                          FILE **fp, const char *ht_buff, size_t ht_len,
                          const char *adr, const char *fil, const char *save);
--- a/src/htscoremain.c
+++ b/src/htscoremain.c
@@ -357,7 +357,7 @@ static int hts_main_internal(int argc, char **argv, httrackp * opt) {
          char BIGSTK tempo[HTS_CDLMAXSIZE];

          strcpybuff(tempo, argv[na] + 1);
-          if (tempo[0] == '\0' || tempo[strlen(tempo) - 1] != '"') {
+          if (hts_lastchar(tempo) != '"') {
            char BIGSTK s[HTS_CDLMAXSIZE];

            sprintf(s, "Missing quote in %s", argv[na]);
@@ -365,7 +365,7 @@ static int hts_main_internal(int argc, char **argv, httrackp * opt) {
            htsmain_free();
            return -1;
          }
-          tempo[strlen(tempo) - 1] = '\0';
+          hts_choplastchar(tempo);
          /* tempo is argv[na] minus its surrounding quotes, so it fits in place
           */
          strlcpybuff(argv[na], tempo, strlen(argv[na]) + 1);
@@ -863,7 +863,7 @@ static int hts_main_internal(int argc, char **argv, httrackp * opt) {
        char BIGSTK tempo[HTS_CDLMAXSIZE + 256];

        strcpybuff(tempo, argv[na] + 1);
-        if (tempo[0] == '\0' || tempo[strlen(tempo) - 1] != '"') {
+        if (hts_lastchar(tempo) != '"') {
          char s[HTS_CDLMAXSIZE + 256];

          sprintf(s, "Missing quote in %s", argv[na]);
@@ -871,7 +871,7 @@ static int hts_main_internal(int argc, char **argv, httrackp * opt) {
          htsmain_free();
          return -1;
        }
-        tempo[strlen(tempo) - 1] = '\0';
+        hts_choplastchar(tempo);
        /* tempo is argv[na] minus its surrounding quotes, so it fits in place
         */
        strlcpybuff(argv[na], tempo, strlen(argv[na]) + 1);
@@ -2724,10 +2724,7 @@ static int hts_main_internal(int argc, char **argv, httrackp * opt) {
      char *a;

      strcpybuff(rpath, StringBuff(opt->path_html));
-      if (rpath[0]) {
-        if (rpath[strlen(rpath) - 1] == '/')
-          rpath[strlen(rpath) - 1] = '\0';
-      }
+      hts_striplastchar(rpath, '/');
      a = strrchr(rpath, '/');
      if (a) {
        *a = '\0';
--- a/src/htsftp.c
+++ b/src/htsftp.c
@@ -624,6 +624,9 @@ int run_launch_ftp(FTPDownloadStruct * pStruct) {
          } else {
            file_notify(opt, back->url_adr, back->url_fil, back->url_sav, 1, 1,
                        0);
+            /* Every failure exit below would else leave the mirror truncated
+               (#771); the resume branch appends and needs no backup. */
+            back_refetch_backup(opt, back);
            back->r.fp = filecreate(&opt->state.strc, back->url_sav);
          }
          strcpybuff(back->info, "receiving");
--- a/src/htshelp.c
+++ b/src/htshelp.c
@@ -83,9 +83,8 @@ void infomsg(const char *msg) {
              /* try the flag as-is, then strip a trailing N as the numeric-arg
                 placeholder (cN -> c); this order keeps -%N from becoming -% */
              p = optreal_find(cmd);
-              if (p < 0 && (int) strlen(cmd) > 2 &&
-                  cmd[strlen(cmd) - 1] == 'N') {
-                cmd[strlen(cmd) - 1] = '\0';
+              if (p < 0 && (int) strlen(cmd) > 2 && hts_lastchar(cmd) == 'N') {
+                hts_striplastchar(cmd, 'N');
                p = optreal_find(cmd);
              }
              if (p >= 0) {
@@ -211,7 +210,7 @@ void help_wizard(httrackp * opt) {
    strcatbuff(str, "/websites/");
  }
  if (strnotempty(str))
-    if ((str[strlen(str) - 1] != '/') && (str[strlen(str) - 1] != '\\'))
+    if ((hts_lastchar(str) != '/') && (hts_lastchar(str) != '\\'))
      strcatbuff(str, "/");
  strcatbuff(stropt2, "-O \"");
  strcatbuff(stropt2, str);
--- a/src/htslib.c
+++ b/src/htslib.c
@@ -5078,7 +5078,7 @@ static int hts_dns_resolve_nocache_list(const char *const hostname,
  if (!strnotempty(hostname) || max <= 0) {
    return 0;
  }
-  if ((hostname[0] == '[') && (hostname[strlen(hostname) - 1] == ']')) {
+  if ((hostname[0] == '[') && (hts_lastchar(hostname) == ']')) {
    size_t size = strlen(hostname);
    char *copy = malloct(size + 1);
    int count;
@@ -5491,9 +5491,8 @@ void cut_path(char *fullpath, char *path, size_t path_size, char *pname,
              size_t pname_size) {
  path[0] = pname[0] = '\0';
  if (strnotempty(fullpath)) {
-    if ((fullpath[strlen(fullpath) - 1] == '/')
-        || (fullpath[strlen(fullpath) - 1] == '\\'))
-      fullpath[strlen(fullpath) - 1] = '\0';
+    if (!hts_striplastchar(fullpath, '/'))
+      hts_striplastchar(fullpath, '\\');
    if (strlen(fullpath) > 1) {
      char *a;

@@ -6648,9 +6647,12 @@ int hts_rename_utf8(const char *oldpath, const char *newpath) {
  LPWSTR wnewpath = hts_pathToUCS2(newpath);
  if (woldpath != NULL && wnewpath != NULL) {
    const int result = _wrename(woldpath, wnewpath);
+    /* Save errno: callers key off it (#779) and free() may clobber it. */
+    const int err = errno;

    free(woldpath);
    free(wnewpath);
+    errno = err;
    return result;
  } else {
    if (woldpath != NULL)
--- a/src/htsname.c
+++ b/src/htsname.c
@@ -437,9 +437,7 @@ int url_savename(lien_adrfilsave *const afs,

      strcpybuff(fil_complete_patche, normfil);
      // Version avec ou sans /
-      if (fil_complete_patche[strlen(fil_complete_patche) - 1] == '/')
-        fil_complete_patche[strlen(fil_complete_patche) - 1] = '\0';
-      else
+      if (!hts_striplastchar(fil_complete_patche, '/'))
        strcatbuff(fil_complete_patche, "/");
      i = hash_read(hash, normadr, fil_complete_patche, HASH_STRUCT_ORIGINAL_ADR_PATH);       // recherche table 2 (former->adr+former->fil)
      if (i >= 0) {
@@ -515,7 +513,8 @@ int url_savename(lien_adrfilsave *const afs,
        && protocol != PROTOCOL_FTP
      ) {
      // tester type avec requète HEAD si on ne connait pas le type du fichier
-      if (!((opt->check_type == 1) && (fil[strlen(fil) - 1] == '/')))   // slash doit être html?
+      if (!((opt->check_type == 1) &&
+            (hts_lastchar(fil) == '/'))) // slash doit être html?
        if (opt->savename_delayed == HTS_SAVENAME_DELAYED_HARD ||
            ishtml(opt, fil) < 0) { // unsure whether it's html or a file
          // lire dans le cache
@@ -810,7 +809,7 @@ int url_savename(lien_adrfilsave *const afs,
  // - - - DEBUT NOMMAGE - - -

  // Donner nom par défaut?
-  if (fil[strlen(fil) - 1] == '/') {
+  if (hts_lastchar(fil) == '/') {
    if (!strfield(adr_complete, "ftp://")
      ) {
      strcatbuff(fil, DEFAULT_HTML);    // nommer page par défaut!!
@@ -1284,7 +1283,7 @@ int url_savename(lien_adrfilsave *const afs,

    hts_lowcase(afs->save);

-    if (afs->save[strlen(afs->save) - 1] == '/')
+    if (hts_lastchar(afs->save) == '/')
      strcatbuff(afs->save, DEFAULT_HTML);   // nommer page par défaut!!
  }

--- a/src/htsparse.c
+++ b/src/htsparse.c
@@ -1701,7 +1701,8 @@ int htsparse(htsmoduleStruct * str, htsmoduleStructExtended * stre) {
 #endif
                                    )   // ok pas de problème
                                    url_ok = 1;
-                                  else if (tempo[strlen(tempo) - 1] == '/') {   // un slash: ok..
+                                  else if (hts_lastchar(tempo) ==
+                                           '/') {       // un slash: ok..
                                    if (inscript)       // sinon si pas javascript, méfiance (répertoire style base?)
                                      url_ok = 1;
                                  }
@@ -2005,9 +2006,8 @@ int htsparse(htsmoduleStruct * str, htsmoduleStructExtended * stre) {
              if (eadr - html - 1 < HTS_URLMAXSIZE) {        // pas trop long?
                strncpy(lien, html, eadr - html - 1);
                lien[eadr - html - 1] = '\0';
-                // supprimer les espaces
-                while((lien[strlen(lien) - 1] == ' ') && (strnotempty(lien)))
-                  lien[strlen(lien) - 1] = '\0';
+                while (hts_striplastchar(lien, ' ')) {
+                }

              } else
                lien[0] = '\0'; // erreur
@@ -2207,7 +2207,7 @@ int htsparse(htsmoduleStruct * str, htsmoduleStructExtended * stre) {
                // supposition dangereuse?
                // OUI!!
 #if HTS_TILDE_SLASH
-                if (lien[strlen(lien) - 1] != '/') {
+                if (hts_lastchar(lien) != '/') {
                  char *a = lien + strlen(lien) - 1;

                  // éviter aussi index~1.html
@@ -2272,7 +2272,7 @@ int htsparse(htsmoduleStruct * str, htsmoduleStructExtended * stre) {
                    // Vérifier les codebase=applet (au lieu de applet/)
                    if (p_type == -2) { // codebase
                      if (strnotempty(lien)) {
-                        if (lien[strlen(lien) - 1] != '/') {     // pas répertoire
+                        if (hts_lastchar(lien) != '/') { // pas répertoire
                          strcatbuff(lien, "/");
                        }
                      }
@@ -2688,9 +2688,11 @@ int htsparse(htsmoduleStruct * str, htsmoduleStructExtended * stre) {
                            int cat_data_len = 0;

                            // ajouter lien external
-                            switch ((link_has_authority(afs.af.adr)) ? 1
-                                    : ((afs.af.fil[strlen(afs.af.fil) - 1] ==
-                                        '/') ? 1 : (ishtml(opt, afs.af.fil)))) {
+                            switch ((link_has_authority(afs.af.adr))
+                                        ? 1
+                                        : ((hts_lastchar(afs.af.fil) == '/')
+                                               ? 1
+                                               : (ishtml(opt, afs.af.fil)))) {
                            case 1:
                            case -2:   // html ou répertoire
                              if (opt->getmode & HTS_GETMODE_HTML) {
@@ -2733,7 +2735,7 @@ int htsparse(htsmoduleStruct * str, htsmoduleStructExtended * stre) {
                                cat_data_len = HTS_DATA_UNKNOWN_HTML_LEN;
                              }
                              break;
-                            }   // html,gif
+                            } // html,gif

                            if (patch_it) {
                              char BIGSTK save[HTS_URLMAXSIZE * 2];
--- a/src/htssafe.h
+++ b/src/htssafe.h
@@ -539,6 +539,36 @@ static HTS_INLINE HTS_UNUSED HTS_PRINTF_FUN(3, 4) void slprintfbuff_clip(
 #define sprintfbuff(ARR, ...) slprintfbuff((ARR), sizeof(ARR), __VA_ARGS__)
 #endif

+/* Last character of s, or '\0' when s is empty. Replaces s[strlen(s) - 1],
+   which indexes one byte before the buffer on an empty string. */
+static HTS_INLINE HTS_UNUSED char hts_lastchar(const char *s) {
+  const size_t len = strlen(s);
+
+  return len != 0 ? s[len - 1] : '\0';
+}
+
+/* Drop a trailing c from s if present; HTS_TRUE if one was dropped. */
+static HTS_INLINE HTS_UNUSED hts_boolean hts_striplastchar(char *s, char c) {
+  const size_t len = strlen(s);
+
+  if (len != 0 && s[len - 1] == c) {
+    s[len - 1] = '\0';
+    return HTS_TRUE;
+  }
+  return HTS_FALSE;
+}
+
+/* Drop the last character of s whatever it is; HTS_TRUE if s was not empty. */
+static HTS_INLINE HTS_UNUSED hts_boolean hts_choplastchar(char *s) {
+  const size_t len = strlen(s);
+
+  if (len != 0) {
+    s[len - 1] = '\0';
+    return HTS_TRUE;
+  }
+  return HTS_FALSE;
+}
+
 /* Thin aliases over the libc allocator/memcpy (historical "t" suffix); no
   added bounds checking. freet() also NULLs the freed pointer and tolerates
   NULL. memcpybuff() despite the name is a raw memcpy: the caller owns the
--- a/src/htsselftest.c
+++ b/src/htsselftest.c
@@ -5955,6 +5955,104 @@ static int st_mirrorio(httrackp *opt, int argc, char **argv) {
  return 0;
 }

+static void ro_put(const char *path, const char *data) {
+  FILE *const fp = FOPEN(path, "wb");
+
+  assertf(fp != NULL);
+  assertf(fwrite(data, 1, strlen(data), fp) == strlen(data));
+  fclose(fp);
+}
+
+/* HTS_TRUE if path holds exactly data. */
+static hts_boolean ro_is(const char *path, const char *data) {
+  char buf[64];
+  FILE *const fp = FOPEN(path, "rb");
+  size_t n;
+
+  if (fp == NULL)
+    return HTS_FALSE;
+  n = fread(buf, 1, sizeof(buf), fp);
+  fclose(fp);
+  return n == strlen(data) && memcmp(buf, data, n) == 0 ? HTS_TRUE : HTS_FALSE;
+}
+
+// -#test=renameover <dir>: hts_rename_over() must replace an existing dst and
+// never delete one it did not replace (#779). Which half is live depends on
+// what rename() does to an existing target, so probe that and name the regime.
+static int st_renameover(httrackp *opt, int argc, char **argv) {
+  (void) opt;
+  if (argc < 1) {
+    fprintf(stderr, "renameover: needs a writable base dir\n");
+    return 1;
+  }
+  char src[HTS_URLMAXSIZE * 2], dst[HTS_URLMAXSIZE * 2];
+  int err = 0;
+
+  fconcat(src, sizeof(src), argv[0], "renameover-src.bin");
+  fconcat(dst, sizeof(dst), argv[0], "renameover-dst.bin");
+
+  (void) UNLINK(src);
+  (void) UNLINK(dst);
+  ro_put(src, "probe");
+  ro_put(dst, "probe");
+
+  const int probe = RENAME(src, dst) == 0 ? 0 : errno;
+  /* Only a target in the way is something the unlink can clear. */
+  const hts_boolean replaceable = probe == 0 || probe == EEXIST;
+
+  printf("renameover: regime %s\n",
+         probe == 0 ? "clobber" : (probe == EEXIST ? "fallback" : "refused"));
+
+  (void) UNLINK(src);
+  (void) UNLINK(dst);
+  ro_put(src, "new");
+  ro_put(dst, "old");
+  if (replaceable) {
+    /* An existing dst must still be replaced: the unlink is for this. */
+    if (!hts_rename_over(src, dst)) {
+      fprintf(stderr, "renameover: replacing an existing dst failed: %s\n",
+              strerror(errno));
+      err++;
+    } else if (!ro_is(dst, "new") || fexist_utf8(src)) {
+      fprintf(stderr, "renameover: dst was not replaced by src\n");
+      err++;
+    }
+  } else {
+    /* A failure the unlink cannot fix must leave dst as it was. */
+    if (hts_rename_over(src, dst)) {
+      fprintf(stderr, "renameover: an unfixable failure reported success\n");
+      err++;
+    }
+    if (!ro_is(dst, "old")) {
+      fprintf(stderr, "renameover: an unfixable failure destroyed dst\n");
+      err++;
+    }
+  }
+
+  /* A missing src must leave dst alone and report failure. */
+  (void) UNLINK(src);
+  ro_put(dst, "keep");
+  if (hts_rename_over(src, dst)) {
+    fprintf(stderr, "renameover: a missing src reported success\n");
+    err++;
+  }
+  if (!ro_is(dst, "keep")) {
+    fprintf(stderr, "renameover: a missing src destroyed dst\n");
+    err++;
+  }
+
+  /* Same, with dst absent too: nothing to lose, still a failure. */
+  (void) UNLINK(dst);
+  if (hts_rename_over(src, dst)) {
+    fprintf(stderr, "renameover: a missing src and dst reported success\n");
+    err++;
+  }
+
+  (void) UNLINK(dst);
+  printf("renameover: %s\n", err ? "FAIL" : "OK");
+  return err;
+}
+
 // -#test=direnum <dir>: enumerate a long+non-ASCII directory via the
 // opendir/readdir wrappers; children must round-trip as UTF-8 (#133,#630).
 static int st_direnum(httrackp *opt, int argc, char **argv) {
@@ -6268,6 +6366,13 @@ static void threadwait_gated_thread(void *arg) {
  hts_mutexrelease(&threadwait_lock);
 }

+static int st_backswap(httrackp *opt, int argc, char **argv) {
+  (void) opt;
+  (void) argc;
+  (void) argv;
+  return back_selftest_slot_swap();
+}
+
 static int st_threadwait(httrackp *opt, int argc, char **argv) {
  int err = 0;
  int i, round;
@@ -6374,7 +6479,7 @@ static int st_changes_race(httrackp *opt, int argc, char **argv) {
    return 1;
  }
  strcpybuff(base, argv[0]);
-  if (base[0] != '\0' && base[strlen(base) - 1] != '/')
+  if (base[0] != '\0' && hts_lastchar(base) != '/')
    strcatbuff(base, "/");
  StringCopy(opt->path_html, base);
  StringCopy(opt->path_html_utf8, base);
@@ -6445,6 +6550,88 @@ static int st_changes_race(httrackp *opt, int argc, char **argv) {
  return err;
 }

+/* The x[strlen(x) - 1] class (#770). The string starts mid-arena so the byte
+   it must not touch is a real neighbour; poisoned with '#', not 0, or a stray
+   NUL terminator would read as untouched. */
+static int st_lastchar(httrackp *opt, int argc, char **argv) {
+  enum { off = 8 };
+
+  char arena[16];
+  char *const s = &arena[off];
+  const int guard = off - 1; /* what the old idiom clobbers */
+  int err = 0;
+
+  (void) opt;
+  (void) argc;
+  (void) argv;
+
+#define REPOISON(str)                                                          \
+  do {                                                                         \
+    memset(arena, '#', sizeof(arena));                                         \
+    strlcpybuff(s, (str), sizeof(arena) - off);                                \
+  } while (0)
+#define CHECK(cond)                                                            \
+  do {                                                                         \
+    if (!(cond)) {                                                             \
+      printf("  FAIL line %d: %s\n", __LINE__, #cond);                         \
+      err = 1;                                                                 \
+    }                                                                          \
+  } while (0)
+
+  /* the empty string: every helper must report "nothing" and touch nothing */
+  REPOISON("");
+  CHECK(hts_lastchar(s) == '\0');
+  CHECK(arena[guard] == '#');
+  REPOISON("");
+  CHECK(hts_striplastchar(s, '/') == HTS_FALSE);
+  CHECK(arena[guard] == '#');
+  CHECK(s[0] == '\0');
+  REPOISON("");
+  CHECK(hts_choplastchar(s) == HTS_FALSE);
+  CHECK(arena[guard] == '#');
+  CHECK(s[0] == '\0');
+
+  /* a '/' sitting where the underflow would land must not be mistaken for the
+     string's own last byte -- this is the #768 shape */
+  REPOISON("");
+  arena[guard] = '/';
+  CHECK(hts_lastchar(s) == '\0');
+  CHECK(hts_striplastchar(s, '/') == HTS_FALSE);
+  CHECK(arena[guard] == '/');
+
+  /* non-empty: ordinary behaviour */
+  REPOISON("ab/");
+  CHECK(hts_lastchar(s) == '/');
+  CHECK(hts_striplastchar(s, '/') == HTS_TRUE);
+  CHECK(strcmp(s, "ab") == 0);
+  CHECK(hts_striplastchar(s, '/') == HTS_FALSE);
+  CHECK(strcmp(s, "ab") == 0);
+  CHECK(hts_choplastchar(s) == HTS_TRUE);
+  CHECK(strcmp(s, "a") == 0);
+  CHECK(hts_choplastchar(s) == HTS_TRUE);
+  CHECK(s[0] == '\0');
+  CHECK(arena[guard] == '#');
+
+  /* one-character string: the boundary the guards get wrong */
+  REPOISON("/");
+  CHECK(hts_lastchar(s) == '/');
+  CHECK(hts_striplastchar(s, '/') == HTS_TRUE);
+  CHECK(s[0] == '\0');
+  CHECK(arena[guard] == '#');
+
+  /* control: the canary must be able to fail, or the checks above prove
+     nothing. Clobber it exactly as the unguarded idiom would. */
+  REPOISON("");
+  s[-1] = '\0';
+  CHECK(arena[guard] != '#');
+
+#undef REPOISON
+#undef CHECK
+
+  printf("lastchar self-test: %s\n", err ? "FAIL" : "OK");
+  return err;
+}
+
 /* ------------------------------------------------------------ */
 /* Registry: name -> handler, with a usage hint and a one-line description. */
 /* ------------------------------------------------------------ */
@@ -6500,12 +6687,16 @@ static const struct selftest_entry {
    {"strsafe", "[overflow|overflow-buff|overflow-src [str]]",
     "bounded string-op self-test", st_strsafe},
    {"copyopt", "", "copy_htsopt option-copy self-test", st_copyopt},
+    {"lastchar", "", "last-char helpers never index before the buffer (#770)",
+     st_lastchar},
    {"changes", "", "--changes bucket accounting and JSON escaping (#714)",
     st_changes},
    {"changes-race", "<dir>", "--changes under a late transfer thread (#714)",
     st_changes_race},
    {"threadwait", "", "htsthread_wait() joins threads spawned just before it",
     st_threadwait},
+    {"backswap", "", "which backlog slots may be swapped to the ready table",
+     st_backswap},
    {"pause", "", "randomized inter-file pause target self-test", st_pause},
    {"relative", "<link> <curr-file>", "relative link between two paths",
     st_relative},
@@ -6603,6 +6794,10 @@ static const struct selftest_entry {
    {"mirrorio", "<dir>",
     "round-trip a long+non-ASCII path through the mirror I/O wrappers",
     st_mirrorio},
+    {"renameover", "<dir>",
+     "hts_rename_over(): replace dst, but never delete a dst it did not "
+     "replace",
+     st_renameover},
    {"direnum", "<dir>",
     "enumerate a long+non-ASCII directory through opendir/readdir",
     st_direnum},
--- a/src/htsserver.c
+++ b/src/htsserver.c
@@ -1050,9 +1050,9 @@ int smallserver(T_SOC soc, char *url, char *method, char *data, char *path) {
                if (!linput(fp, line, sizeof(line) - 2)) {
                  *str = '\0';
                }
-                if (*str && str[strlen(str) - 1] == '\\') {
+                if (hts_lastchar(str) == '\\') {
                  nocr = 1;
-                  str[strlen(str) - 1] = '\0';
+                  hts_striplastchar(str, '\\');
                }
                while(*str) {
                  char *pos;
@@ -1169,11 +1169,8 @@ int smallserver(T_SOC soc, char *url, char *method, char *data, char *path) {
                          char *rpath = (char *) adr;

                          //find_handle h;
-                          if (rpath[0]) {
-                            if (rpath[strlen(rpath) - 1] == '/') {
-                              rpath[strlen(rpath) - 1] = '\0';  /* note: patching stored (inhash) value */
-                            }
-                          }
+                          /* note: patching stored (inhash) value */
+                          hts_striplastchar(rpath, '/');
                          {
                            const char *profiles = hts_getcategories(rpath, 0);
                            const char *categ = hts_getcategories(rpath, 1);
--- a/src/htstools.c
+++ b/src/htstools.c
@@ -985,10 +985,7 @@ HTSEXT_API int hts_buildtopindex(httrackp * opt, const char *path,
      && toptemplate_bodycat) {

    strcpybuff(rpath, path);
-    if (rpath[0]) {
-      if (rpath[strlen(rpath) - 1] == '/')
-        rpath[strlen(rpath) - 1] = '\0';
-    }
+    hts_striplastchar(rpath, '/');

    fpo = fopen(fconcat(catbuff, sizeof(catbuff), rpath, "/index.html"), "wb");
    if (fpo) {
@@ -1201,11 +1198,8 @@ HTSEXT_API char *hts_getcategories(char *path, int type) {
  find_handle h;
  coucal hashCateg = NULL;

-  if (rpath[0]) {
-    if (rpath[strlen(rpath) - 1] == '/') {
-      rpath[strlen(rpath) - 1] = '\0';  /* note: patching stored (inhash) value */
-    }
-  }
+  /* note: patching stored (inhash) value */
+  hts_striplastchar(rpath, '/');
  h = hts_findfirst(rpath);
  if (h) {
    String iname = STRING_EMPTY;
@@ -1302,7 +1296,7 @@ HTSEXT_API find_handle hts_findfirst(char *path) {

          strcpybuff(rpath, path);
          if (rpath[0]) {
-            if (rpath[strlen(rpath) - 1] != '\\')
+            if (hts_lastchar(rpath) != '\\')
              strcatbuff(rpath, "\\");
          }
          strcatbuff(rpath, "*.*");
@@ -1314,7 +1308,7 @@ HTSEXT_API find_handle hts_findfirst(char *path) {
        strcpybuff(find->path, path);
        {
          if (find->path[0]) {
-            if (find->path[strlen(find->path) - 1] != '/')
+            if (hts_lastchar(find->path) != '/')
              strcatbuff(find->path, "/");
          }
        }
@@ -1451,7 +1445,14 @@ hts_boolean hts_rename_over(const char *src, const char *dst) {
  fconv(cdst, sizeof(cdst), dst);
  if (RENAME(csrc, cdst) == 0)
    return HTS_TRUE;
-  /* RENAME does not clobber an existing target on Windows. */
+  /* Only a dst in the way is something the unlink can clear, and the CRT maps
+     that to EEXIST; it keeps EACCES for a src another process holds, where
+     removing dst would lose a file the retry cannot replace (#790). The src
+     check covers a CRT that reports neither. */
+  const int err = errno;
+
+  if (err != EEXIST || !fexist_utf8(src))
+    return HTS_FALSE;
  (void) UNLINK(cdst);
  return RENAME(csrc, cdst) == 0 ? HTS_TRUE : HTS_FALSE;
 }
--- a/src/htstools.h
+++ b/src/htstools.h
@@ -138,7 +138,9 @@ HTSEXT_API hts_boolean hts_findisfile(find_handle find);
 HTSEXT_API hts_boolean hts_findissystem(find_handle find);

 /* Move src onto dst, replacing an existing dst; HTS_TRUE on success. Both
-   paths are fconv()'d. */
+   paths are fconv()'d. dst is removed only to make room for a src that exists,
+   so a caller whose src was never written keeps its dst; a retry that still
+   fails does not (#790). */
 hts_boolean hts_rename_over(const char *src, const char *dst);

 #endif
--- a/src/htswarc.c
+++ b/src/htswarc.c
@@ -57,6 +57,9 @@ Please visit our Website: http://www.httrack.com
 /* opt->state.warc value meaning "open failed once, do not retry". */
 #define WARC_DISABLED ((void *) ~(uintptr_t) 0)

+/* Suffix of the in-progress archive when a previous one must survive it. */
+#define WARC_TMP_SUFFIX ".tmp"
+
 struct warc_writer {
  FILE *f;
  httrackp *opt;   /* kept for close-time logging (warc_wacz_package) */
@@ -86,6 +89,11 @@ struct warc_writer {
  char *base_path;      /* resolved archive path minus .warc[.gz] suffix */
  const char *base_ext; /* ".warc.gz" or ".warc" (static) */
  char *arc_path;    /* full single-file archive path (NULL under rotation) */
+  /* A re-run must not destroy an archive it cannot replace (#759). */
+  hts_boolean protect_prev;   /* previous archive present: build in a temp */
+  hts_boolean opened;         /* open completed; a failed one swaps nothing */
+  hts_boolean failed;         /* a record or segment was lost: swap nothing */
+  uint64_t unbacked_revisits; /* revisits whose payload no file here holds */
  char **page_lines; /* one JSON page line per 200 text/html response, owned */
  size_t page_count;
  size_t page_cap;
@@ -1129,13 +1137,17 @@ static int warc_emit(warc_writer *w, const char *type, const char *content_type,
     never split a record, and never rotate a warcinfo (it opens a segment). */
  if (w->max_size > 0 && w->seg_base != NULL && w->offset >= w->max_size &&
      strcmp(type, "warcinfo") != 0) {
-    if (warc_rotate(w) != 0)
+    if (warc_rotate(w) != 0) {
+      w->failed = HTS_TRUE;
      return -1;
+    }
  }

  /* F4: overflow-safe block length; http_hdr_len+sep is provably small. */
-  if (payload > (size_t) -1 - http_hdr_len - sep)
+  if (payload > (size_t) -1 - http_hdr_len - sep) {
+    w->failed = HTS_TRUE;
    return -1;
+  }
  block_len = http_hdr_len + sep + payload;

  memset(&hdr, 0, sizeof(hdr));
@@ -1241,11 +1253,23 @@ static int warc_emit(warc_writer *w, const char *type, const char *content_type,
  rc = 0;
 done:
  wbuf_free(&hdr);
+  if (rc != 0)
+    w->failed = HTS_TRUE; /* a truncated run must not replace a whole one */
  return rc;
 }

 /* ---- segment rotation (--warc-max-size) ---- */

+/* Path to open for the segment whose final path is `final`: that path itself,
+   or a sibling temp while a previous archive must survive until close. */
+static const char *warc_open_path(warc_writer *w, const char *final, char *buf,
+                                  size_t bufsz) {
+  if (!w->protect_prev)
+    return final;
+  snprintf(buf, bufsz, "%s" WARC_TMP_SUFFIX, final);
+  return buf;
+}
+
 /* Emit the warcinfo that heads a segment; sets w->info_id for its records. */
 static int warc_write_warcinfo_record(warc_writer *w) {
  w->info_id[0] = '\0'; /* warcinfo itself carries no WARC-Warcinfo-ID */
@@ -1257,17 +1281,23 @@ static int warc_write_warcinfo_record(warc_writer *w) {

 static int warc_rotate(warc_writer *w) {
  char namebuf[HTS_URLMAXSIZE * 2];
+  char openbuf[HTS_URLMAXSIZE * 2 + sizeof(WARC_TMP_SUFFIX)];
  char catbuff[CATBUFF_SIZE];
-  if (w->f != NULL) {
-    fclose(w->f);
-    w->f = NULL;
-  }
-  w->seg++;
-  snprintf(namebuf, sizeof(namebuf), "%s-%05u%s", w->seg_base, w->seg,
+  const unsigned next = w->seg + 1;
+  FILE *f;
+  snprintf(namebuf, sizeof(namebuf), "%s-%05u%s", w->seg_base, next,
           w->seg_ext);
-  w->f = FOPEN(fconv(catbuff, sizeof(catbuff), namebuf), "wb");
-  if (w->f == NULL)
+  /* Open before advancing: w->seg must only ever name a segment that exists,
+     or close-time packaging and swapping would work on a missing file. */
+  f = FOPEN(fconv(catbuff, sizeof(catbuff),
+                  warc_open_path(w, namebuf, openbuf, sizeof(openbuf))),
+            "wb");
+  if (f == NULL)
    return -1;
+  if (w->f != NULL)
+    fclose(w->f);
+  w->f = f;
+  w->seg = next;
  w->offset = 0;
  if (w->cdx_on) {
    freet(w->cur_seg);
@@ -1324,6 +1354,7 @@ void warc_adopt_rawspool(htsblk *r, const char *tmpfile_path) {
 warc_writer *warc_open(httrackp *opt, const char *path) {
  warc_writer *w;
  char namebuf[HTS_URLMAXSIZE * 2];
+  char openbuf[HTS_URLMAXSIZE * 2 + sizeof(WARC_TMP_SUFFIX)];
  char catbuff[CATBUFF_SIZE];
  wbuf info;
  const char *robots;
@@ -1474,35 +1505,105 @@ warc_writer *warc_open(httrackp *opt, const char *path) {
    path = namebuf;
  }

-  w->f = FOPEN(fconv(catbuff, sizeof(catbuff), path), "wb");
+  if (w->max_size == 0 && (w->arc_path = strdupt(path)) == NULL) {
+    warc_close(w);
+    return NULL;
+  }
+  /* Set only once arc_path is recorded, so a half-built writer never swaps. */
+  w->protect_prev = fsize_utf8(path) > 0 ? HTS_TRUE : HTS_FALSE;
+  w->f = FOPEN(fconv(catbuff, sizeof(catbuff),
+                     warc_open_path(w, path, openbuf, sizeof(openbuf))),
+               "wb");
  if (w->f == NULL) {
    warc_close(w);
    return NULL;
  }
  if (w->cdx_on)
    w->cur_seg = path_basename_dup(path);
-  if (w->wacz_on && w->max_size == 0)
-    w->arc_path = strdupt(path); /* single-file: package this exact path */

  if (warc_write_warcinfo_record(w) != 0) {
    warc_close(w);
    return NULL;
  }
+  w->opened = HTS_TRUE;
  return w;
 }

+/* Final path of segment s (the run's only archive when rotation is off). */
+static const char *warc_seg_path(warc_writer *w, unsigned s, char *buf,
+                                 size_t bufsz) {
+  if (w->max_size == 0)
+    return w->arc_path;
+  snprintf(buf, bufsz, "%s-%05u%s", w->seg_base, s, w->seg_ext);
+  return buf;
+}
+
+/* Swap this run's archive into place, unless it only holds revisits naming
+   bodies the previous archive still has and this one does not (#759).
+   HTS_FALSE: the previous archive was kept, so leave its .cdx and .wacz too. */
+static hts_boolean warc_commit(warc_writer *w) {
+  char finalbuf[HTS_URLMAXSIZE * 2];
+  char tmpbuf[HTS_URLMAXSIZE * 2 + sizeof(WARC_TMP_SUFFIX)];
+  char catbuff[CATBUFF_SIZE];
+  const unsigned nseg = (w->max_size > 0) ? w->seg + 1 : 1;
+  hts_boolean swap;
+  unsigned s;
+
+  if (!w->protect_prev)
+    return HTS_TRUE; /* nothing was there to lose: written in place */
+
+  /* All or nothing: a swap stopping halfway would mix this run's segments with
+     the previous one's, so require every temp before renaming any. */
+  swap = w->opened && !w->failed && w->unbacked_revisits == 0;
+  for (s = 0; s < nseg && swap; s++) {
+    snprintf(tmpbuf, sizeof(tmpbuf), "%s" WARC_TMP_SUFFIX,
+             warc_seg_path(w, s, finalbuf, sizeof(finalbuf)));
+    swap = fsize_utf8(tmpbuf) > 0;
+  }
+
+  if (swap) {
+    for (s = 0; s < nseg; s++) {
+      const char *final = warc_seg_path(w, s, finalbuf, sizeof(finalbuf));
+      snprintf(tmpbuf, sizeof(tmpbuf), "%s" WARC_TMP_SUFFIX, final);
+      if (!hts_rename_over(tmpbuf, final)) {
+        hts_log_print(w->opt, LOG_ERROR | LOG_ERRNO,
+                      "WARC: could not replace %s", final);
+        return HTS_FALSE;
+      }
+    }
+    return HTS_TRUE;
+  }
+
+  for (s = 0; s < nseg; s++) {
+    snprintf(tmpbuf, sizeof(tmpbuf), "%s" WARC_TMP_SUFFIX,
+             warc_seg_path(w, s, finalbuf, sizeof(finalbuf)));
+    (void) UNLINK(fconv(catbuff, sizeof(catbuff), tmpbuf));
+  }
+  if (w->unbacked_revisits > 0)
+    hts_log_print(
+        w->opt, LOG_ERROR,
+        "WARC: this pass revisited %llu URL(s) without re-downloading them, so "
+        "its archive would name bodies no file holds; kept the previous %s "
+        "(re-run with -C0, or --warc-file with a name of its own)",
+        (unsigned long long) w->unbacked_revisits,
+        warc_seg_path(w, 0, finalbuf, sizeof(finalbuf)));
+  return HTS_FALSE;
+}
+
 void warc_close(warc_writer *w) {
  size_t i;
  if (w == NULL)
    return;
-  warc_cdx_flush(w); /* sort + write <base>.cdx before tearing down */
  if (w->f != NULL)
    fclose(w->f);
  w->f = NULL;
+  if (warc_commit(w)) {
+    warc_cdx_flush(w); /* sort + write <base>.cdx beside the archive */
 #if HTS_USEOPENSSL
-  if (w->wacz_on) /* package once the segment(s) + .cdx are closed on disk */
-    warc_wacz_package(w);
+    if (w->wacz_on) /* package once the segment(s) + .cdx are closed on disk */
+      warc_wacz_package(w);
 #endif
+  }
  if (w->seen != NULL)
    coucal_delete(&w->seen);
  for (i = 0; i < w->cdx_count; i++)
@@ -1583,6 +1684,8 @@ int warc_write_transaction(warc_writer *w, const char *target_uri,
  if (is_update_unchanged) {
    is_revisit = 1;
    profile = "http://netpreserve.org/warc/1.1/revisit/server-not-modified";
+    /* Served from cache: the payload sits in the previous archive, not here. */
+    w->unbacked_revisits++;
  } else if (have_pdig && w->seen != NULL) {
    void *prev = NULL;
    if (coucal_read_pvoid(w->seen, pdig, &prev) && prev != NULL) {
--- a/tests/01_engine-backswap.test
+++ b/tests/01_engine-backswap.test
@@ -0,0 +1,8 @@
+#!/bin/bash
+#
+
+set -euo pipefail
+
+# A ready slot still owning a temporary must stay in memory: swapping it out
+# clears the entry, which unlinks the re-fetch backup (#771).
+httrack -O /dev/null -#test=backswap | grep -q "backswap self-test: OK"
--- a/tests/01_engine-lastchar.test
+++ b/tests/01_engine-lastchar.test
@@ -0,0 +1,35 @@
+#!/bin/bash
+#
+# Runs the lastchar self-test (#770), then keeps the idiom it replaced from
+# coming back: x[strlen(x) - 1] indexes one byte before the buffer when x is
+# empty, and the guard is never where the index is.
+
+set -eu
+
+: "${top_srcdir:=..}"
+
+out=$(httrack -#test=lastchar) || {
+    echo "httrack -#test=lastchar exited non-zero: $out" >&2
+    exit 1
+}
+
+# Exact-match the success line so a renamed/removed test can't pass silently.
+test "$out" = "lastchar self-test: OK" || {
+    echo "expected 'lastchar self-test: OK', got: $out" >&2
+    exit 1
+}
+
+# The self-test only covers the helpers, so guard the class at the source: the
+# lone survivor is inside a comment block in htsname.c. coucal is vendored, so
+# its idioms are not ours to police.
+hits=$(command grep -rn 'strlen([^)]*) *- *1\]' "$top_srcdir/src" \
+    --include='*.c' --include='*.h' --exclude-dir=coucal |
+    grep -v 'htsname\.c:.*save\[strlen(save)-1\]' |
+    grep -v 'Replaces s\[strlen(s) - 1\]' |
+    grep -v 'The x\[strlen(x) - 1\] class' || true)
+
+test -z "$hits" || {
+    echo "x[strlen(x) - 1] reintroduced; use hts_lastchar/hts_striplastchar/hts_choplastchar:" >&2
+    echo "$hits" >&2
+    exit 1
+}
--- a/tests/01_engine-renameover.test
+++ b/tests/01_engine-renameover.test
@@ -0,0 +1,55 @@
+#!/bin/bash
+#
+
+set -euo pipefail
+
+# Drives -#test=renameover: hts_rename_over() must replace an existing dst, and
+# must leave dst alone when the rename failed for a reason removing dst cannot
+# fix (#779). The selftest prints the regime it detected; pin it per platform so
+# a leg cannot pass having tested the other half.
+dir=$(mktemp -d)
+trap 'rm -rf "$dir"' EXIT
+
+case "$(uname -s)" in
+MINGW* | MSYS_NT*) want=fallback ;; # native rename() refuses an existing target
+*) want=clobber ;;
+esac
+
+out=$(httrack -O /dev/null -#test=renameover "$dir")
+echo "$out" | grep -q "renameover: OK"
+echo "$out" | grep -q "renameover: regime $want"
+
+if [ "$(uname -s)" != "Linux" ]; then
+    echo "renameover: LD_PRELOAD interposition is Linux-only here, skipping"
+    exit 0
+fi
+# A --disable-shared build has nothing to preload; anything else missing is a
+# build problem, not a skip, or the interposed legs would pass vacuously.
+if [ ! -r "${RENAMEFAIL_LA:-}" ]; then
+    echo "renameover: ${RENAMEFAIL_LA:-\$RENAMEFAIL_LA} was not built" >&2
+    exit 1
+fi
+if grep -q "^dlname=''" "$RENAMEFAIL_LA"; then
+    echo "renameover: static-only build, skipping the interposed legs"
+    exit 0
+fi
+if [ ! -r "${RENAMEFAIL_LIB:-}" ]; then
+    echo "renameover: ${RENAMEFAIL_LIB:-\$RENAMEFAIL_LIB} was not built" >&2
+    exit 1
+fi
+
+# An LD_PRELOAD library loads ahead of the executable's own libasan, which ASan
+# refuses by default. The shim allocates nothing, so the ordering is harmless.
+export ASAN_OPTIONS="${ASAN_OPTIONS:+$ASAN_OPTIONS:}verify_asan_link_order=0"
+
+# The unlink fallback is dead code on POSIX, so borrow Windows' rename().
+out=$(LD_PRELOAD="$RENAMEFAIL_LIB" httrack -O /dev/null \
+    -#test=renameover "$dir")
+echo "$out" | grep -q "renameover: OK"
+echo "$out" | grep -q "renameover: regime fallback"
+
+# A source another process holds fails with EACCES, which dst had no part in.
+out=$(RENAMEFAIL_MODE=locked LD_PRELOAD="$RENAMEFAIL_LIB" httrack -O /dev/null \
+    -#test=renameover "$dir")
+echo "$out" | grep -q "renameover: OK"
+echo "$out" | grep -q "renameover: regime refused"
--- a/tests/100_local-purge-longpath.test
+++ b/tests/100_local-purge-longpath.test
@@ -0,0 +1,111 @@
+#!/bin/bash
+#
+# The end-of-mirror purge read old.lst with linput(fp, line, 1000), which caps a
+# chunk at 999 bytes. filenote() writes "[<save path>]", so a 998-byte save path
+# makes a 1000-byte line that comes back as 999 bytes plus a one-byte "]" tail.
+# For that tail `line + 1` is empty, and with no -O so is path_html, so the purge
+# indexed file[-1] (#770). Needs no -O, hence not local-crawl.sh.
+#
+# The OOB write lands in stack padding, so only the sanitized CI leg fails on
+# unfixed code; the length assertion below is what keeps this from passing
+# vacuously if the save path ever stops landing on the chunk boundary.
+
+set -euo pipefail
+
+: "${top_srcdir:=..}"
+
+# shellcheck source=tests/testlib.sh
+. "$top_srcdir/tests/testlib.sh"
+
+# python3 is required; mirror check-network.sh's skip-with-77 convention.
+python=$(find_python) || ! echo "python3 not found; skipping" >&2 || exit 77
+
+# Resolve httrack now: the harness prepends a RELATIVE dir to PATH, so the cd
+# below would otherwise silently fall through to an installed copy.
+httrack_bin=$(command -v httrack)
+test -n "$httrack_bin" || {
+    echo "httrack not on PATH" >&2
+    exit 1
+}
+httrack_bin=$(cd "$(dirname "$httrack_bin")" && pwd)/$(basename "$httrack_bin")
+
+work=$(mktemp -d)
+trap 'kill -9 "${spid:-}" 2>/dev/null || true; rm -rf "$work"' EXIT
+
+mkdir -p "$work/root" "$work/proj"
+
+"$python" "$top_srcdir/tests/local-server.py" --root "$work/root" \
+    >"$work/server.out" 2>"$work/server.err" &
+spid=$!
+
+port=
+for _ in $(seq 1 100); do
+    port=$(sed -n 's/^PORT \([0-9]*\)$/\1/p' "$work/server.out" 2>/dev/null || true)
+    test -n "$port" && break
+    sleep 0.1
+done
+test -n "$port" || {
+    echo "local-server.py did not announce a port" >&2
+    exit 1
+}
+
+# Save path is "127.0.0.1_<port>/" + the URL path; target 998 bytes total.
+prefix="127.0.0.1_${port}/"
+need=$((998 - ${#prefix}))
+
+# macOS caps PATH_MAX at 1024, so neither the docroot nor the mirror can hold a
+# 998-byte path, and httrack's own HTS_MAX_PATH_LEN clamps below it too. The
+# trigger is simply not reachable there.
+pathmax=$(getconf PATH_MAX / 2>/dev/null || echo 4096)
+test "$pathmax" -ge $((need + ${#work} + 128)) || {
+    echo "PATH_MAX $pathmax too small for a $need-byte path; skipping" >&2
+    exit 77
+}
+
+# Same on Win32, where MAX_PATH clamps the save path; getconf answers for the
+# MSYS emulation, not the Win32 API httrack calls, so it cannot see this.
+case "$(uname -s)" in
+MINGW* | MSYS* | CYGWIN*)
+    echo "Win32 MAX_PATH cannot hold a $need-byte save path; skipping" >&2
+    exit 77
+    ;;
+esac
+
+urlpath=""
+while test $((need - ${#urlpath})) -gt 60; do
+    urlpath="${urlpath}$(printf 'd%02d' ${#urlpath})$(printf 'a%.0s' $(seq 1 46))/"
+done
+urlpath="${urlpath}$(printf 'f%.0s' $(seq 1 $((need - ${#urlpath} - 5)))).html"
+test ${#urlpath} -eq "$need"
+
+mkdir -p "$work/root/$(dirname "$urlpath")"
+printf '<html><body>leaf</body></html>\n' >"$work/root/$urlpath"
+printf '<html><body><a href="/%s">x</a></body></html>\n' "$urlpath" \
+    >"$work/root/index.html"
+
+cd "$work/proj" # no -O: opt->path_html stays empty
+
+"$httrack_bin" "http://127.0.0.1:$port/" -q -%v0 -s0 -r3 >"$work/crawl1.log" 2>&1 || true
+
+lst=$work/proj/hts-cache/new.lst
+test -f "$lst" || {
+    echo "pass 1 produced no new.lst" >&2
+    exit 1
+}
+# Arms the test: without a line of exactly 1000 bytes linput() never splits and
+# the purge never sees a one-byte tail.
+awk 'length == 1000 { found = 1 } END { exit !found }' "$lst" || {
+    echo "no 1000-byte line in new.lst; the chunk split is not exercised" >&2
+    awk '{ print length }' "$lst" | sort -n | tail -3 >&2
+    echo "--- urlpath len ${#urlpath}, prefix $prefix" >&2
+    tail -20 "$work/crawl1.log" >&2
+    exit 1
+}
+
+# Pass 2 renames new.lst to old.lst and runs it through the purge loop.
+"$httrack_bin" "http://127.0.0.1:$port/" -q -%v0 -s0 -r3 >"$work/crawl2.log" 2>&1
+
+test -s "$work/proj/$prefix$urlpath" || {
+    echo "the mirrored leaf did not survive the purge" >&2
+    exit 1
+}
--- a/tests/101_local-update-stale-bak.test
+++ b/tests/101_local-update-stale-bak.test
@@ -0,0 +1,27 @@
+#!/bin/bash
+#
+# A .bak left behind by a killed run must not disable the #77 re-fetch backup
+# (#758): a leftover file (slow.bin.bak) has to be clobbered so the -M abort can
+# still restore the pass-1 copy, and a leftover the engine cannot clobber
+# (fast.bin.bak, planted as a directory) has to be reported instead of silently
+# truncating with no safety net. Only the Windows leg arms the first half:
+# POSIX rename() clobbers on its own.
+
+set -euo pipefail
+
+: "${top_srcdir:=..}"
+
+bash "$top_srcdir/tests/local-crawl.sh" \
+    --plant-file bigtrunc/slow.bin.bak \
+    --plant-dir bigtrunc/fast.bin.bak \
+    --rerun-args '--update -M400000' \
+    --log-found 'More than 400000 bytes have been transferred.. giving up' \
+    --log-found 'replacing leftover backup .*slow\.bin\.bak' \
+    --log-found 'could not back up .*fast\.bin' \
+    --log-not-found 'could not back up .*slow\.bin' \
+    --log-not-found 'could not restore' \
+    --found bigtrunc/slow.bin \
+    --found bigtrunc/fast.bin \
+    --file-min-bytes bigtrunc/slow.bin 655360 \
+    --file-not-matches bigtrunc/slow.bin 'stale-leftover' \
+    httrack 'BASEURL/bigtrunc/index.html' -c2
--- a/tests/102_local-ftp-refetch.test
+++ b/tests/102_local-ftp-refetch.test
@@ -0,0 +1,163 @@
+#!/bin/bash
+#
+# An FTP re-fetch used to truncate the mirror before the transfer, so a failed
+# one destroyed the previous copy (#771). keep.bin is cut short and empty.bin
+# gets nothing; both must keep their old bytes. stay.bin takes the same path and
+# completes, so a fix that merely stopped writing would fail here.
+
+set -euo pipefail
+
+: "${top_srcdir:=..}"
+testdir=$(cd "$(dirname "$0")" && pwd)
+# shellcheck source=tests/testlib.sh
+. "${testdir}/testlib.sh"
+
+python=$(find_python) || ! echo "python3 not found; skipping" >&2 || exit 77
+command -v httrack >/dev/null || {
+    echo "could not find httrack" >&2
+    exit 1
+}
+
+server=$(nativepath "${testdir}/ftp-server.py")
+tmpdir=$(mktemp -d "${TMPDIR:-/tmp}/httrack_ftp.XXXXXX")
+serverpid=
+cleanup() {
+    stop_server "$serverpid"
+    rm -rf "$tmpdir"
+}
+trap cleanup EXIT HUP INT QUIT PIPE TERM
+
+root="${tmpdir}/root"
+out="${tmpdir}/crawl"
+mode="${tmpdir}/mode"
+report="${out}/hts-changes.json"
+mkdir -p "$root" "$out"
+
+# The two generations differ in length, or a re-fetch that resumed at EOF
+# instead of truncating would leave the same bytes and pass for the wrong reason.
+write_bodies() {
+    local gen="$1" fill="$2" name
+    for name in keep empty stay; do
+        "$python" -c 'import sys; sys.stdout.buffer.write(
+            ("%s-FTP-%s " % (sys.argv[1].upper(), sys.argv[2])).encode()
+            + sys.argv[2][-1].encode() * int(sys.argv[3]))' "$name" "$gen" "$fill" \
+            >"${root}/${name}.bin"
+    done
+}
+write_bodies V1 4096
+: >"$mode"
+
+serverlog="${tmpdir}/server.out"
+"$python" "$server" --root "$(nativepath "$root")" \
+    --mode-file "$(nativepath "$mode")" >"$serverlog" 2>&1 &
+serverpid=$!
+port=
+for _ in $(seq 1 300); do
+    line=$(grep -m1 '^PORT ' "$serverlog" 2>/dev/null) && port="${line#PORT }" && break
+    kill -0 "$serverpid" 2>/dev/null || {
+        echo "ftp server exited early: $(cat "$serverlog")" >&2
+        exit 1
+    }
+    sleep 0.1
+done
+test -n "$port" || {
+    echo "could not discover ftp server port: $(cat "$serverlog")" >&2
+    exit 1
+}
+host="127.0.0.1_${port}"
+urls=()
+for name in keep empty stay; do
+    urls+=("ftp://127.0.0.1:${port}/${name}.bin")
+done
+# -c1: a parallel FTP crawl loses whole transfers to a separate, older bug in
+# the backlog slot swap (#797), which would flake this test on a loaded runner.
+common=(--quiet --disable-security-limits --robots=0 --timeout=20 --max-time=120
+    --retries=1 -c1 --changes)
+
+fail() {
+    echo "FAIL: $*" >&2
+    test ! -f "$report" || cat "$report" >&2
+    exit 1
+}
+ok() { echo "OK: $*"; }
+size_of() { wc -c <"$1" | tr -d '[:space:]'; }
+# Files listed under a report bucket, one "file:size" per line.
+listed() {
+    "$python" - "$report" "$1" <<'EOF' | tr -d '\r'
+import json
+import sys
+
+with open(sys.argv[1], encoding="utf-8") as fp:
+    report = json.load(fp)
+for entry in report[sys.argv[2]]:
+    print("%s:%s" % (entry["file"], entry.get("size", "none")))
+EOF
+}
+
+# --- pass 1: a healthy mirror ------------------------------------------------
+# Bounded like every local-crawl.sh pass: a wedged crawl must fail the test, not
+# hang the job until CI cancels it and discards the log (#796).
+run_with_timeout 300 httrack "${urls[@]}" -O "$out" "${common[@]}" \
+    >"${tmpdir}/log1" 2>&1
+for name in keep empty stay; do
+    test -s "${out}/${host}/${name}.bin" || fail "pass 1 did not mirror ${name}.bin"
+done
+mkdir "${tmpdir}/snap"
+cp "${out}/${host}"/*.bin "${tmpdir}/snap/"
+ok "pass 1 mirrored three files"
+
+# --- pass 2: the same three re-fetched, two of them failing -------------------
+# norest throughout: an accepted REST sends the client down the append path,
+# which is not the one that truncates.
+write_bodies V2 6000
+cat >"$mode" <<'EOF'
+/keep.bin truncate norest
+/empty.bin empty norest
+/stay.bin norest
+EOF
+run_with_timeout 300 httrack "${urls[@]}" -O "$out" "${common[@]}" --update \
+    >"${tmpdir}/log2" 2>&1
+
+# Both failures have to be the transfer dying mid-body: a connect or lookup
+# failure never opens the file, and everything below would pass vacuously.
+for name in keep empty; do
+    grep -aqE "Error:.*\"FTP file incomplete\".*${name}\.bin" "${out}/hts-log.txt" ||
+        fail "${name}.bin did not fail during the transfer"
+done
+ok "both re-fetches reached the body and failed there"
+
+for name in keep empty; do
+    cmp -s "${out}/${host}/${name}.bin" "${tmpdir}/snap/${name}.bin" ||
+        fail "${name}.bin differs from the copy pass 1 left ($(size_of \
+            "${out}/${host}/${name}.bin") bytes now)"
+done
+ok "a failed FTP transfer left the previously mirrored bytes alone"
+
+grep -aq "FTP-V2 " "${out}/${host}/stay.bin" ||
+    fail "stay.bin was not refreshed; the re-fetch writes nothing at all now"
+ok "a successful FTP re-fetch still replaces the file"
+
+leftover=$(find "${out}/${host}" -name '*.bak')
+test -z "$leftover" || fail "backup left behind: ${leftover}"
+ok "no backup temporary survived the pass"
+
+# --purge-old is on by default: a kept file the run never replaced must still be
+# in new.lst, or it is deleted as gone (#746).
+for name in keep empty stay; do
+    test -f "${out}/${host}/${name}.bin" || fail "${name}.bin was purged"
+done
+ok "the kept copies survived the update purge"
+
+# --- the change report is the mirror's own account of the pass ----------------
+test -s "$report" || fail "pass 2 wrote no ${report}"
+test -z "$(listed gone)" || fail "the report calls something gone: $(listed gone)"
+for name in keep empty; do
+    listed unchanged | grep -qx "${host}/${name}.bin:$(size_of "${tmpdir}/snap/${name}.bin")" ||
+        fail "${name}.bin is not reported unchanged at its previous size"
+    if listed changed | grep -q "^${host}/${name}.bin:"; then
+        fail "${name}.bin is reported changed as well"
+    fi
+done
+listed changed | grep -qx "${host}/stay.bin:$(size_of "${out}/${host}/stay.bin")" ||
+    fail "stay.bin is not reported changed"
+ok "the change report calls the kept files unchanged and the refreshed one changed"
--- a/tests/37_local-cache-outage.test
+++ b/tests/37_local-cache-outage.test
@@ -1,12 +1,14 @@
 #!/bin/bash
 #
 # An update run against a dead server must not destroy the cache: the no-data
-# rollback restores the previous hts-cache generation (zip caches lost it).
+# rollback restores the previous hts-cache generation (zip caches lost it). The
+# live update pass leaves an old.zip behind, so the dead pass rotates onto an
+# existing generation instead of into empty space.

 set -eu

 : "${top_srcdir:=..}"

-bash "$top_srcdir/tests/local-crawl.sh" --errors 0 --rerun-dead \
+bash "$top_srcdir/tests/local-crawl.sh" --errors 0 --rerun --rerun-dead \
    --found 'simple/basic.html' \
    httrack 'BASEURL/simple/basic.html'
--- a/tests/73_local-warc.test
+++ b/tests/73_local-warc.test
@@ -1,11 +1,14 @@
 #!/bin/bash
 #
-# A --warc crawl writes a standards-conformant WARC/1.1 archive, and an
-# --update re-crawl of an unchanged (all-304) site emits revisit records.
-# The stdlib validator (no warcio) is the real gate: it byte-compares the
-# fresh page.html response body against what the server served, asserts the
-# encoding headers were stripped and the payload digest matches, then checks
-# the update pass turned the unchanged assets into revisits (no full response).
+# A --warc crawl writes a standards-conformant WARC/1.1 archive, and a re-crawl
+# of an unchanged (all-304) site emits revisit records. The stdlib validator (no
+# warcio) is the real gate: it byte-compares the fresh page.html response body
+# against what the server served, asserts the encoding headers were stripped and
+# the payload digest matches, then checks the second pass turned the unchanged
+# assets into revisits (no full response).
+#
+# That second pass archives under a name of its own, which is also the recovery
+# the engine points at when it refuses to strand a previous archive (#759).

 set -eu

@@ -14,8 +17,10 @@ set -eu
 # page.html body served by tests/local-server.py (route_mini304_page).
 export WARC_VALIDATE_BODY="page.html=3c68746d6c3e3c626f64793e74696e7920636163686561626c6520706167653c2f626f64793e3c2f68746d6c3e0a"
 export WARC_VALIDATE_NORESP="index.html page.html"
+export WARC_VALIDATE_UPDATE="warc-update.warc.gz"

-bash "$top_srcdir/tests/local-crawl.sh" --errors 0 --rerun --warc-validate \
+bash "$top_srcdir/tests/local-crawl.sh" --errors 0 --warc-validate \
+    --rerun-args '--warc-file warc-update' \
    --log-found 'no files updated' \
    --found 'mini304/index.html' --found 'mini304/page.html' \
    httrack 'BASEURL/mini304/index.html' --warc-file warc-out
--- a/tests/74_local-warc-wacz.test
+++ b/tests/74_local-warc-wacz.test
@@ -6,9 +6,10 @@
 # the datapackage-digest chain (plus py-wacz/pywb when importable). The crawl
 # skips cleanly on a build without OpenSSL (no conformant SHA-256 -> no package).
 #
-# The cacheless second pass repackages over the .wacz the first one left: the
-# clobber, not the happy rename, is what breaks when a platform refuses to
-# rename onto an existing file (#726).
+# The cacheless (-C0) second pass repackages over the .wacz the first one left:
+# the clobber, not the happy rename, is what breaks when a platform refuses to
+# rename onto an existing file (#726). A cache-served pass keeps the previous
+# package instead, which is 97_local-warc-update-keep.

 set -eu

--- a/tests/96_local-refetch-keep.test
+++ b/tests/96_local-refetch-keep.test
@@ -1,19 +1,16 @@
 #!/bin/bash
 #
-# A re-fetch cut mid-header must leave the mirrored file alone (#748): unfixed,
-# the aborted read's leftover buffer lands on it. err.bin is the control, an
-# HTTP 500 on the same resource already keeping its copy; stay.bin answers
-# normally with a new body, so a fix that stopped overwriting anything fails.
-#
-# Purging is off: an update purge deletes both files for an unrelated reason
-# (#746), which would hide what this asserts.
+# A re-fetch cut mid-header must leave the mirrored file alone (#748) and keep
+# it out of the update purge (#746). err.bin is the control, an HTTP 500 on the
+# same resource already surviving both; stay.bin answers normally with a new
+# body, so a fix that stopped overwriting anything fails.

 set -euo pipefail

 : "${top_srcdir:=..}"

 bash "$top_srcdir/tests/local-crawl.sh" \
-    --rerun-args '--update --purge-old=0' \
+    --rerun-args '--update' \
    --log-found 'after 2 retries at link .*keep/page\.html' \
    --log-found 'after 2 retries at link .*keep/data\.bin' \
    --file-matches 'keep/page.html' 'KEEP-PAGE-V1' \
--- a/tests/97_local-warc-update-keep.test
+++ b/tests/97_local-warc-update-keep.test
@@ -0,0 +1,15 @@
+#!/bin/bash
+#
+# A cache-served second pass fetches no bodies, so it must leave the WARC, CDXJ
+# and WACZ of the first pass untouched (#759). The kept package is re-validated
+# to prove it still replays, not just that a same-named file is still there.
+# The one Error: line is the keep notice; a second means the pass also failed.
+
+set -eu
+
+: "${top_srcdir:=..}"
+
+bash "$top_srcdir/tests/local-crawl.sh" --errors 1 --archive-kept-on-rerun \
+    --rerun --wacz-validate --log-found 'WARC: this pass revisited' \
+    --found 'mini304/index.html' --found 'mini304/page.html' \
+    httrack 'BASEURL/mini304/index.html' --warc-file warc-out --wacz
--- a/tests/98_local-warc-segments.test
+++ b/tests/98_local-warc-segments.test
@@ -0,0 +1,24 @@
+#!/bin/bash
+#
+# The same keep/replace rule under --warc-max-size, where the archive is several
+# segments: a cache-served pass must leave every one alone, a cacheless pass must
+# swap every one in. Half a swap is the corruption one rename loop can produce,
+# and only a rotated archive exposes it (#759). --archive-min-files pins the two
+# segments plus the .cdx, so a shrinking crawl cannot quietly stop rotating.
+
+set -eu
+
+: "${top_srcdir:=..}"
+
+# The one Error: line is the keep notice; a second means the pass also failed.
+bash "$top_srcdir/tests/local-crawl.sh" --errors 1 --archive-kept-on-rerun \
+    --archive-min-files 3 --rerun --log-found 'WARC: this pass revisited' \
+    --found 'mini304/index.html' --found 'mini304/page.html' \
+    httrack 'BASEURL/mini304/index.html' --warc-file warc-out --warc-cdx \
+    --warc-max-size 600
+
+bash "$top_srcdir/tests/local-crawl.sh" --errors 0 --archive-replaced-on-rerun \
+    --archive-min-files 3 --rerun-args '-C0' \
+    --found 'mini304/index.html' --found 'mini304/page.html' \
+    httrack 'BASEURL/mini304/index.html' --warc-file warc-out --warc-cdx \
+    --warc-max-size 600
--- a/tests/99_local-robots-error.test
+++ b/tests/99_local-robots-error.test
@@ -0,0 +1,17 @@
+#!/bin/bash
+# Issue #769: a failed robots.txt probe under -o1, with .txt assumed image/gif,
+# still drives the --sitemap fallback and substitutes no placeholder body.
+
+set -eu
+
+: "${top_srcdir:=..}"
+
+# -Z: both the removed block's own log line and the failed probe that gates
+# this test's coverage are debug-level.
+bash "$top_srcdir/tests/local-crawl.sh" --errors 0 \
+    --found 'sitemapdir/orphan1.html' \
+    --log-found 'No robots\.txt rules at' \
+    --log-found 'listed by 127\.0\.0\.1:[0-9]+/sitemap\.xml' \
+    --log-not-found 'Creating GIF dummy file' \
+    httrack 'BASEURL/sitemapdir/start.html' --sitemap -r2 -o1 -Z \
+    --assume 'txt=image/gif' -F 'errorrobots-agent'
--- a/tests/Makefile.am
+++ b/tests/Makefile.am
@@ -1,11 +1,12 @@
 # Committed binary fixture read by 01_zlib-cache-golden.test. List it
 # explicitly: automake does not expand wildcards in EXTRA_DIST, so a glob would
 # silently drop it from the dist tarball and break "make distcheck".
-EXTRA_DIST = $(TESTS) crawl-test.sh run-all-tests.sh check-network.sh \
+EXTRA_DIST = $(TESTS) renamefail.c crawl-test.sh run-all-tests.sh check-network.sh \
 	proxy-https-server.py socks5-server.py proxy-connect-server.py \
 	proxytestlib.py tls-stall-server.py warc-validate.py wacz-validate.py \
 	pty-resize.py \
-	local-crawl.sh local-server.py testlib.sh server.crt server.key \
+	local-crawl.sh local-server.py ftp-server.py testlib.sh \
+	server.crt server.key \
 	server-root/simple/basic.html server-root/simple/link.html \
 	server-root/stripquery/index.html server-root/stripquery/a.html \
 	server-root/fraglink/index.html server-root/fraglink/target.html \
@@ -21,6 +22,15 @@ TESTS_ENVIRONMENT += BROTLI_ENABLED=$(BROTLI_ENABLED)
 TESTS_ENVIRONMENT += ZSTD_ENABLED=$(ZSTD_ENABLED)
 TESTS_ENVIRONMENT += V6_SUPPORT=$(V6_SUPPORT)
 TESTS_ENVIRONMENT += top_srcdir=$(top_srcdir)
+TESTS_ENVIRONMENT += RENAMEFAIL_LA=$(abs_builddir)/librenamefail.la
+TESTS_ENVIRONMENT += RENAMEFAIL_LIB=$(abs_builddir)/$(LT_CV_OBJDIR)/librenamefail.so
+
+# rename() interposer for 01_engine-renameover.test; -rpath is what makes
+# libtool build a shared module rather than a static-only convenience library.
+check_LTLIBRARIES = librenamefail.la
+librenamefail_la_SOURCES = renamefail.c
+librenamefail_la_LDFLAGS = -module -avoid-version -rpath $(abs_builddir)
+librenamefail_la_LIBADD = $(DL_LIBS)

 TEST_EXTENSIONS = .test
 # Run each .test through bash instead of execve()ing it. This lets "make check"
@@ -49,6 +59,7 @@ TESTS = \
 	01_engine-filterdual.test \
 	01_engine-ftp-line.test \
 	01_engine-ftp-userpass.test \
+	01_engine-backswap.test \
 	01_engine-cacheindex.test \
 	01_engine-hashtable.test \
 	01_engine-header.test \
@@ -59,6 +70,7 @@ TESTS = \
 	01_engine-identabs.test \
 	01_engine-escape-room.test \
 	01_engine-inplace-escape.test \
+	01_engine-lastchar.test \
 	01_engine-makeindex.test \
 	01_engine-mime.test \
 	01_engine-parse.test \
@@ -74,6 +86,7 @@ TESTS = \
 	01_engine-direnum.test \
 	01_engine-cookieimport.test \
 	01_engine-relative.test \
+	01_engine-renameover.test \
 	01_engine-robots.test \
 	01_engine-savename.test \
 	01_engine-selftest-dispatch.test \
@@ -198,6 +211,12 @@ TESTS = \
 	93_local-changes.test \
 	94_local-single-file.test \
 	95_local-sitemap.test \
-	96_local-refetch-keep.test
+	96_local-refetch-keep.test \
+	97_local-warc-update-keep.test \
+	98_local-warc-segments.test \
+	99_local-robots-error.test \
+	100_local-purge-longpath.test \
+	101_local-update-stale-bak.test \
+	102_local-ftp-refetch.test

 CLEANFILES = check-network_sh.cache
--- a/tests/ftp-server.py
+++ b/tests/ftp-server.py
@@ -0,0 +1,227 @@
+#!/usr/bin/env python3
+"""Minimal FTP server for the crawl tests: PASV only, binary RETR, LIST.
+
+Prints "PORT <n>" once bound, like local-server.py. --mode-file is re-read
+before every transfer, so a test can flip one path's behaviour between passes
+without restarting the server and moving the port (which names the mirror
+directory). Each line is "<path> <mode>...", path "*" matching everything:
+
+  truncate  send a prefix of the body, then drop the data connection
+  empty     open the data connection and send nothing
+  norest    answer REST with 500, so the client re-fetches from scratch
+"""
+
+import argparse
+import os
+import socket
+import sys
+import threading
+
+
+def reply(conn, text):
+    conn.sendall((text + "\r\n").encode("utf-8", "replace"))
+
+
+class Session(threading.Thread):
+    def __init__(self, conn, root, mode_file, log):
+        threading.Thread.__init__(self, daemon=True)
+        self.conn = conn
+        self.root = root
+        self.mode_file = mode_file
+        self.log = log
+        self.pasv = None
+        self.rest = 0
+        self.path = "/"  # named by SIZE/RETR; REST carries no path
+
+    def modes(self):
+        if not self.mode_file:
+            return set()
+        found = set()
+        try:
+            with open(self.mode_file, encoding="utf-8") as fp:
+                for line in fp:
+                    words = line.split()
+                    if words and words[0] in ("*", self.path):
+                        found |= set(words[1:])
+        except OSError:
+            pass
+        return found
+
+    # Resolve an FTP path argument under the root, refusing escapes.
+    def resolve(self, arg):
+        arg = arg.strip().strip('"')
+        self.path = "/" + arg.lstrip("/")
+        path = os.path.normpath(os.path.join(self.root, arg.lstrip("/")))
+        if path != self.root and not path.startswith(self.root + os.sep):
+            return None
+        return path
+
+    def open_pasv(self):
+        srv = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
+        srv.bind(("127.0.0.1", 0))
+        srv.listen(1)
+        self.pasv = srv
+        port = srv.getsockname()[1]
+        return "227 Entering Passive Mode (127,0,0,1,%d,%d)" % (port >> 8, port & 0xFF)
+
+    def accept_data(self):
+        if self.pasv is None:
+            return None
+        self.pasv.settimeout(30)
+        try:
+            data, _ = self.pasv.accept()
+        except (OSError, socket.timeout):
+            data = None
+        self.pasv.close()
+        self.pasv = None
+        return data
+
+    def send_body(self, body, modes):
+        data = self.accept_data()
+        if data is None:
+            reply(self.conn, "425 no data connection")
+            return
+        try:
+            if "empty" in modes:
+                pass
+            elif "truncate" in modes:
+                data.sendall(body[: max(1, len(body) // 8)])
+            else:
+                data.sendall(body)
+        except OSError:
+            pass
+        data.close()
+        if modes & {"empty", "truncate"}:
+            reply(self.conn, "426 transfer aborted")
+        else:
+            reply(self.conn, "226 Transfer complete")
+
+    def do_retr(self, arg):
+        path = self.resolve(arg)
+        if path is None or not os.path.isfile(path):
+            reply(self.conn, "550 no such file")
+            return
+        with open(path, "rb") as fp:
+            body = fp.read()
+        modes = self.modes()
+        reply(self.conn, "150 opening binary mode")
+        self.send_body(body[self.rest :], modes)
+        self.rest = 0
+
+    def do_list(self, arg):
+        if arg.startswith("-A"):
+            arg = arg[2:].strip()
+        path = self.resolve(arg or "/")
+        if path is None or not os.path.isdir(path):
+            reply(self.conn, "550 no such directory")
+            return
+        lines = []
+        for name in sorted(os.listdir(path)):
+            full = os.path.join(path, name)
+            if os.path.isdir(full):
+                lines.append("drwxr-xr-x 2 ftp ftp 4096 Jan 01 00:00 %s" % name)
+            else:
+                lines.append(
+                    "-rw-r--r-- 1 ftp ftp %d Jan 01 00:00 %s"
+                    % (os.path.getsize(full), name)
+                )
+        reply(self.conn, "150 opening ASCII mode")
+        self.send_body(("\r\n".join(lines) + "\r\n").encode("utf-8"), set())
+
+    def dispatch(self, verb, arg):
+        conn = self.conn
+        if verb in ("USER", "PASS", "TYPE", "NOOP"):
+            reply(conn, "200 ok")
+        elif verb == "SYST":
+            reply(conn, "215 UNIX Type: L8")
+        elif verb == "PWD":
+            reply(conn, '257 "/"')
+        elif verb == "CWD":
+            reply(conn, "250 ok")
+        elif verb == "PASV":
+            reply(conn, self.open_pasv())
+        elif verb == "EPSV":
+            reply(conn, "500 not supported")
+        elif verb == "SIZE":
+            path = self.resolve(arg)
+            if path and os.path.isfile(path):
+                reply(conn, "213 %d" % os.path.getsize(path))
+            else:
+                reply(conn, "550 no such file")
+        elif verb == "REST":
+            if "norest" in self.modes():
+                reply(conn, "500 REST not understood")
+            else:
+                try:
+                    self.rest = int(arg)
+                except ValueError:
+                    self.rest = 0
+                reply(conn, "350 restarting")
+        elif verb == "RETR":
+            self.do_retr(arg)
+        elif verb in ("LIST", "NLST"):
+            self.do_list(arg)
+        elif verb == "QUIT":
+            reply(conn, "221 bye")
+            return False
+        else:
+            reply(conn, "500 unknown command")
+        return True
+
+    def run(self):
+        conn = self.conn
+        buf = b""
+        try:
+            reply(conn, "220 httrack test ftp")
+            while True:
+                while b"\r\n" not in buf:
+                    chunk = conn.recv(4096)
+                    if not chunk:
+                        return
+                    buf += chunk
+                line, buf = buf.split(b"\r\n", 1)
+                line = line.decode("utf-8", "replace")
+                self.log(line)
+                verb, _, arg = line.partition(" ")
+                if not self.dispatch(verb.upper(), arg):
+                    return
+        except OSError:
+            return
+        finally:
+            if self.pasv is not None:
+                self.pasv.close()
+            conn.close()
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--root", required=True)
+    ap.add_argument("--mode-file")
+    ap.add_argument("--log")
+    args = ap.parse_args()
+
+    logfp = open(args.log, "a", encoding="utf-8") if args.log else None
+    log_lock = threading.Lock()
+
+    def log(line):
+        if logfp:
+            with log_lock:
+                logfp.write(line + "\n")
+                logfp.flush()
+
+    srv = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
+    srv.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
+    srv.bind(("127.0.0.1", 0))
+    srv.listen(16)
+    sys.stdout.reconfigure(newline="\n")  # the launcher parses PORT, CRLF breaks it
+    sys.stdout.write("PORT %d\n" % srv.getsockname()[1])
+    sys.stdout.flush()
+
+    root = os.path.abspath(args.root)
+    while True:
+        conn, _ = srv.accept()
+        Session(conn, root, args.mode_file, log).start()
+
+
+if __name__ == "__main__":
+    main()
--- a/tests/local-crawl.sh
+++ b/tests/local-crawl.sh
@@ -34,6 +34,15 @@
 # httrack via --cookies-file, to exercise preloaded cookies.
 # --rerun-dead re-runs with the server stopped: the no-data rollback must
 # restore the previous hts-cache generation byte-identical.
+# --archive-kept-on-rerun: the second pass must leave the first pass's
+# .warc[.gz]/.cdx/.wacz byte-identical, having no bodies to replace them (#759).
+# --archive-replaced-on-rerun is its mirror: every one of them must have been
+# rewritten. Both also require no *.tmp left behind, and take an optional
+# --archive-min-files N guarding against a scenario that silently stopped
+# producing the segments it means to check.
+# --plant-file/--plant-dir drop a regular file (holding $plant_poison) or a
+# directory at PATH under the host root between the passes, to hand the second
+# pass leftovers a killed run would have left (#758).

 set -u

@@ -54,11 +63,15 @@ outdir_intl=
 rerun=
 rerun_args=
 rerun_dead=
+archive_kept=
+archive_replaced=
+archive_min_files=0
 tmpdir=
 serverpid=
 crawlpid=
 wacz_poisoned=
 wacz_poison="stale-wacz-that-a-second-pass-must-replace"
+plant_poison="stale-leftover-that-a-second-pass-must-clobber"

 function warning {
    echo "** $*" >&2
@@ -87,6 +100,18 @@ function cleanup {
    fi
 }

+hostroot=
+function find_hostroot {
+    local cand
+    for cand in "${mirrorroot}/127.0.0.1_${port}" "${mirrorroot}/127.0.0.1"; do
+        if test -d "$cand"; then
+            hostroot="$cand"
+            return 0
+        fi
+    done
+    die "could not find host root under $out"
+}
+
 function assert_equals {
    info "$1"
    if test ! "$2" == "$3"; then
@@ -109,6 +134,7 @@ tmpdir=$(mktemp -d "${tmptopdir}/httrack_local.XXXXXX") || die "could not create
 # --- parse leading control flags --------------------------------------------
 declare -a audit=()
 declare -a cookies=()
+declare -a plants=()
 scheme=http
 pos=0
 args=("$@")
@@ -118,6 +144,13 @@ while test "$pos" -lt "$nargs"; do
    --debug) verbose=1 ;;
    --rerun) rerun=1 ;;           # run httrack a second time (update pass) before auditing
    --rerun-dead) rerun_dead=1 ;; # re-run with the server stopped (cache rollback)
+    # the second pass must leave the first pass's archive files untouched
+    --archive-kept-on-rerun) archive_kept=1 ;;
+    --archive-replaced-on-rerun) archive_replaced=1 ;; # ...or rewrite all of them
+    --archive-min-files)
+        pos=$((pos + 1))
+        archive_min_files="${args[$pos]}"
+        ;;
    # validate the produced .warc.gz (see the validation block near the end)
    --warc-validate) warc_validate=1 ;;
    # validate the produced .wacz package (stdlib, plus py-wacz/pywb if present)
@@ -141,6 +174,10 @@ while test "$pos" -lt "$nargs"; do
        pos=$((pos + 1))
        cookies+=("${args[$pos]}")
        ;;
+    --plant-file | --plant-dir)
+        plants+=("${args[$pos]}" "${args[$((pos + 1))]}")
+        pos=$((pos + 1))
+        ;;
    --rerun-args)
        pos=$((pos + 1))
        rerun_args="${args[$pos]}"
@@ -274,14 +311,47 @@ if test -n "$warc_validate"; then
    test -z "$w1" || cp "$w1" "${tmpdir}/warc-pass1.gz"
 fi

+# Snapshot the archive files the second pass must keep (or must replace).
+declare -a kept_files=()
+if test -n "${archive_kept}${archive_replaced}"; then
+    while read -r f; do
+        test -n "$f" || continue
+        cp "$f" "${tmpdir}/kept-${#kept_files[@]}" || die "could not snapshot $f"
+        kept_files+=("$f")
+    done < <(find "$mirrorroot" -maxdepth 2 \
+        \( -name '*.warc.gz' -o -name '*.warc' -o -name '*.cdx' -o -name '*.wacz' \) \
+        2>/dev/null | sort)
+    test "${#kept_files[@]}" -gt 0 ||
+        die "the first pass produced no archive to compare against"
+    test "${#kept_files[@]}" -ge "$archive_min_files" ||
+        die "only ${#kept_files[@]} archive file(s), wanted $archive_min_files: ${kept_files[*]}"
+fi
+
 # Poison the first-pass .wacz when a second pass follows: repackaging moves the
 # new archive over it, so the marker must be gone afterwards (#726). Poisoning
 # beats comparing the two packages, which can come out byte-identical.
-if test -n "$wacz_validate" && test -n "${rerun}${rerun_args}"; then
+if test -z "$archive_kept" && test -n "$wacz_validate" && test -n "${rerun}${rerun_args}"; then
    wacz_poisoned=$(find "$mirrorroot" -maxdepth 2 -name '*.wacz' 2>/dev/null | sort | tail -n1)
    test -z "$wacz_poisoned" || echo "$wacz_poison" >"$wacz_poisoned"
 fi

+# --- plant leftovers the second pass has to deal with ------------------------
+if test "${#plants[@]}" -gt 0; then
+    find_hostroot
+    i=0
+    while test "$i" -lt "${#plants[@]}"; do
+        path="${hostroot}/${plants[$((i + 1))]}"
+        info "planting ${plants[$i]} ${plants[$((i + 1))]}"
+        if test "${plants[$i]}" = "--plant-dir"; then
+            mkdir -p "$path" || die "could not create $path"
+        else
+            echo "$plant_poison" >"$path" || die "could not write $path"
+        fi
+        result "OK"
+        i=$((i + 2))
+    done
+fi
+
 # --- optional second pass: re-mirror into the same dir (cache/update path) ----
 if test -n "$rerun"; then
    info "re-running httrack (update pass)"
@@ -326,6 +396,29 @@ if test -n "$rerun_args"; then
    result "OK (second pass)"
 fi

+# --- optional: did the second pass keep, or replace, the whole archive? ------
+if test "${#kept_files[@]}" -gt 0; then
+    i=0
+    for f in "${kept_files[@]}"; do
+        if test -n "$archive_kept"; then
+            info "checking the second pass kept $(basename "$f")"
+            cmp -s "${tmpdir}/kept-${i}" "$f" ||
+                die "$(basename "$f") was rewritten: the previous archive was destroyed"
+        else
+            info "checking the second pass replaced $(basename "$f")"
+            cmp -s "${tmpdir}/kept-${i}" "$f" &&
+                die "$(basename "$f") still holds the first pass's bytes"
+        fi
+        result "OK"
+        i=$((i + 1))
+    done
+    # A leftover in-progress file is as bad: the next pass would silently eat it.
+    info "checking no in-progress archive was left behind"
+    leftover=$(find "$mirrorroot" -maxdepth 2 \( -name '*.warc.gz.tmp' -o -name '*.warc.tmp' \) 2>/dev/null | head -n1)
+    test -z "$leftover" || die "left behind $leftover"
+    result "OK"
+fi
+
 # --- optional dead pass: server stopped, the cache must survive the rollback --
 if test -n "$rerun_dead"; then
    zip="${out}/hts-cache/new.zip"
@@ -363,14 +456,7 @@ if test -n "$rerun_dead"; then
 fi

 # --- discover the single host root (127.0.0.1_<port> or 127.0.0.1) -----------
-hostroot=
-for cand in "${mirrorroot}/127.0.0.1_${port}" "${mirrorroot}/127.0.0.1"; do
-    if test -d "$cand"; then
-        hostroot="$cand"
-        break
-    fi
-done
-test -n "$hostroot" || die "could not find host root under $out"
+find_hostroot
 debug "host root: $hostroot"

 # --- optional WARC validation (stdlib validator, no warcio) ------------------
@@ -397,14 +483,18 @@ if test -n "$warc_validate"; then
        die "fresh WARC validation failed"
    result "OK"

-    # Final file: after an update pass the unchanged assets must be revisits.
-    if test -n "$rerun"; then
+    # After an update pass the unchanged assets must be revisits, in an archive
+    # of the pass's own (WARC_VALIDATE_UPDATE): a revisit-only pass never
+    # replaces the archive holding the bodies it would strand (#759).
+    if test -n "${WARC_VALIDATE_UPDATE:-}"; then
+        upd=$(find "$mirrorroot" -maxdepth 2 -name "$WARC_VALIDATE_UPDATE" 2>/dev/null | head -n1)
+        test -n "$upd" || die "no $WARC_VALIDATE_UPDATE produced under $mirrorroot"
        declare -a revargs=(--expect-revisit)
        for sub in ${WARC_VALIDATE_NORESP:-}; do
            revargs+=(--no-response-for "$sub")
        done
        info "validating update WARC (revisits)"
-        "$python" "$validator" "$(nativepath "$warc")" "${revargs[@]}" >&2 ||
+        "$python" "$validator" "$(nativepath "$upd")" "${revargs[@]}" >&2 ||
            die "update WARC validation failed"
        result "OK"
    fi
--- a/tests/local-server.py
+++ b/tests/local-server.py
@@ -555,10 +555,22 @@ class Handler(SimpleHTTPRequestHandler):
    # ... and one that points the sitemap at the site root, to check a subtree
    # crawl is not widened by where the site chooses to put its sitemap.
    SCOPE_SITEMAP_UA = "scopesitemap"
+    # ... and one whose probe fails, with a body past the 1070 bytes that
+    # issue #769's over-read needed.
+    ERROR_ROBOTS_UA = "errorrobots"

    def route_robots(self):
        # The Sitemap: record is group-independent; only --sitemap acts on it.
        ua = self.headers.get("User-Agent") or ""
+        if self.ERROR_ROBOTS_UA in ua:
+            body = b"# " + b"x" * 4000 + b"\n"
+            self.send_response(404, "Not Found")
+            self.send_header("Content-Type", "text/plain")
+            self.send_header("Content-Length", str(len(body)))
+            self.end_headers()
+            if self.command != "HEAD":
+                self.wfile.write(body)
+            return
        host = self.headers.get("Host")
        body = "User-agent: *\nDisallow:\n"
        if self.DENY_DECLARED_UA in ua:
--- a/tests/renamefail.c
+++ b/tests/renamefail.c
@@ -0,0 +1,41 @@
+/* Borrows Windows' rename() for 01_engine-renameover.test, since POSIX cannot
+   reach hts_rename_over()'s unlink fallback: an existing target is refused with
+   EEXIST, and RENAMEFAIL_MODE=locked reports EACCES instead, as the CRT does
+   for a source another process holds. */
+
+#define _GNU_SOURCE
+#include <dlfcn.h>
+#include <errno.h>
+#include <stddef.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/stat.h>
+
+/* The tree builds with -fvisibility=hidden, which would hide the interposer. */
+#define SHIM_EXPORT __attribute__((visibility("default")))
+
+SHIM_EXPORT int rename(const char *oldpath, const char *newpath);
+
+SHIM_EXPORT int rename(const char *oldpath, const char *newpath) {
+  static int (*real_rename)(const char *, const char *) = NULL;
+  const char *const mode = getenv("RENAMEFAIL_MODE");
+  const int locked = mode != NULL && strcmp(mode, "locked") == 0;
+  struct stat st;
+
+  if (locked) {
+    errno = EACCES;
+    return -1;
+  }
+  if (stat(newpath, &st) == 0) {
+    errno = EEXIST;
+    return -1;
+  }
+  if (real_rename == NULL) {
+    *(void **) &real_rename = dlsym(RTLD_NEXT, "rename");
+    if (real_rename == NULL) {
+      errno = ENOSYS;
+      return -1;
+    }
+  }
+  return real_rename(oldpath, newpath);
+}
Author	SHA1	Message	Date
Xavier Roche	5751991897	Merge origin/master after #784 ; bound test 102's crawls Every other network crawl runs under a deadline through local-crawl.sh. This test drives its own FTP fixture directly, so a wedge would hang the job until CI cancels it and discards the log (#796). Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Xavier Roche <roche@httrack.com>	2026-07-27 12:17:46 +02:00
Xavier Roche	95f8ebaa44	hts_rename_over() deletes its destination when the source is missing (#784 ) The unlink-then-rename fallback in `hts_rename_over()` is there because Windows' rename() refuses an existing target, but it fired on any failed rename, ENOENT on the source included. A caller moving a temp file it never managed to write lost the destination and got `HTS_FALSE` back, which reads as "nothing happened". #754 unified four hand-rolled copies into this helper, so every call site inherited it. The unlink now runs only for EEXIST, and only with a source that exists. EEXIST is the value that matters: the CRT maps ERROR_ALREADY_EXISTS there and keeps EACCES for a source another process holds, so accepting EACCES as well would have deleted the destination for a failure it had no part in, and the retry would then fail with the destination already gone. The source check is belt and braces for a CRT that reports neither. `hts_rename_utf8()` preserves errno across its free() calls now, since the gate reads it. The existing call sites all derive their source's existence from a create that had to succeed first, so the bug was latent rather than live, but three of those destinations are user data: a mirrored file, a finished .wacz, and a rewritten page nothing will re-fetch. What is left of the window after this is #790. The selftest probes what rename() does to an existing target and asserts against the regime it finds, so it runs three ways on Linux: native, then under an LD_PRELOAD rename() with Windows' shape, then under one reporting a locked source. The harness pins the expected regime per platform, so no leg can pass having exercised the other half. Seven mutants of the gate were each confirmed to fail it. Closes #779 Signed-off-by: Xavier Roche <roche@httrack.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-27 12:07:12 +02:00
Xavier Roche	3a096c589a	x[strlen(x) - 1] indexes before the buffer on an empty string (#780 ) Three bugs have already come out of the same one-liner: `x[strlen(x) - 1]` indexes one byte before the buffer when the string turns out to be empty (#729, #730, #768). Rather than wait for a fourth, this replaces the idiom everywhere with helpers that cannot underflow. All 48 occurrences were reclassified by tracing each guard to where it actually lives instead of to the two lines above the index. 44 were already guarded, often a dozen lines up or in the caller, and one sits inside a commented-out block. The rest have no guard at the site, and two of those are reachable from a crawl. `htscore.c:2194` is the one that matters: a one-byte stack out-of-bounds write in the end-of-mirror purge, confirmed under ASan and UBSan. `linput()` reads `old.lst` in 999-byte chunks, so a 1000-byte line comes back as 999 bytes plus a one-byte tail; for that tail `line + 1` is empty, and without `-O` so is `path_html`, which leaves `file` empty when the index runs. The crawled site controls the save-path length and therefore the line length. It needs a second run over the same project, which is what a re-crawl does. As controls, a 900-byte path and the same path with `-O` both stay clean. `htsparse.c:2009` is a one-byte overread in the HTML parser, reachable from `<a href=" ">`, where the guard sits on the wrong side of the `&&`. The write on the next line is accidentally safe for the same reason. #768 itself turned out not to be reachable from a crawl, only through the `-#test=savename` hook, so I would not call that one a security fix. The helpers live in `htssafe.h` beside the other bounded string operations, so no file needed a new include. Guards doing more than an emptiness check are kept, since folding those in would append a separator to an empty string. `-#test=lastchar` checks the helpers against a poisoned neighbouring byte, including the `/`-before-the-buffer case from #768, and putting the missing length check back makes it fail. It also greps the source, so reverting any converted site fails instead of leaving the self-test green. `97_local-purge-longpath` drives the purge site end to end; on unfixed code the sanitized CI leg aborts at `htscore.c:2194`. Built from identical source paths, 61 of 72 objects are byte-identical, one differs only in debug info, and the remaining 10 are the files edited. The pointer spelling `x + strlen(x) - 1` has 25 occurrences and is filed separately as #781. Closes #770 Closes #768 Signed-off-by: Xavier Roche <roche@httrack.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-27 11:44:43 +02:00
Xavier Roche	1a4a5082b7	The --errpage placeholder block has been unreachable since 3.20.2 (#783 ) The stand-in body `httpmirror()` builds for an error page has been unreachable since the 3.20.2 import: both guards say the opposite of their comments, so the block runs only when there is no save name and the URL is `/robots.txt`. `create_html_warning` is never assigned, so the HTML arm is dead outright. The GIF arm fires only if the user maps `.txt` to `image/gif` with `--assume`, and it then swaps `r.adr` for a 1070-byte buffer while leaving `r.size` at the error body's length, so the `robots_parse()` call below over-reads the heap. Repair is not the one-character fix the inverted guards suggest. `filesave()` below writes `r.size` bytes, so uninverting the guards without also setting `r.size` moves the same over-read into the user's mirror. A correct repair would then replace every server error body with HTTrack's 2003 template by default, which is what `23_local-errpage.test` asserts is kept. Nobody has asked for the placeholder in twenty years, and #17 asked for the opposite. `--errpage` itself is untouched: `store_errpage` keeps the server's error body and the normal save writes it. A byte-level differential against a master build over five error shapes under both `-o1` and `-o0` gives identical mirrors, and rebuilt objects differ only in `htscore.o`. The new test drives the one input that reached the block, and fails on master in a plain build rather than only under the sanitizers. Separately, the `-o` help text promises a generated page the engine has never produced; that predates this change and is filed as #787. Closes #769 Signed-off-by: Xavier Roche <roche@httrack.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-27 11:31:47 +02:00
Xavier Roche	5378fe529a	Merge remote-tracking branch 'origin/master' into fix/ftp-refetch-truncates Signed-off-by: Xavier Roche <roche@httrack.com> # Conflicts: # src/htsback.c	2026-07-27 11:15:13 +02:00
Xavier Roche	d7312da326	tests: renumber the FTP re-fetch test past master Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Xavier Roche <roche@httrack.com>	2026-07-27 11:13:17 +02:00
Xavier Roche	1ad59352f2	Do not swap out a backlog slot that still owns a temporary back_cleanup_background() moves a ready slot to the on-disk table via back_clear_entry(), which unlinks back->tmpfile. A failed re-fetch sits at STATUS_READY unfinalized, so its .bak was deleted before back_finalize() could put it back, losing the previous copy about half the time under load. Covered by -#test=backswap, the crawl-level race needing concurrency the suite cannot pin down. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Xavier Roche <roche@httrack.com>	2026-07-27 11:12:22 +02:00
Xavier Roche	4fd6767a8f	A stale .bak silently disables the re-fetch backup on Windows (#776 ) `back_finalize()` moves an existing file aside to `<file>.bak` before truncating it, so an aborted re-fetch can put the previous copy back. That rename was a bare `RENAME`, and Windows refuses to rename onto an existing target, so a `.bak` outliving a killed run made every later re-fetch of that URL fail the rename, take the `tmpfile = NULL` branch with nothing logged, and truncate the live file with no backup at all. The #77 guard was off for that file, silently and for good. `hts_rename_over()` (#754) unlinks and retries, which fixes it. Worth a look, because clobbering a stale `.bak` is not free. A run killed between the rename-aside and the finalize leaves the previous complete copy in `.bak` and a partial in the live file, and the next re-fetch now overwrites the good one. It still looks like the right trade: POSIX `rename()` has behaved exactly this way since #77 landed, so the alternative is leaving Windows with a guard that stays dead until someone deletes the file by hand, and a leftover `.bak` is engine garbage that nothing advertises or ever restores. The clobber is logged now, so it is at least visible. Any failure to create the backup is logged too, instead of quietly disabling the safety net. `htscache.c`'s static `hts_rename()` wrapper and the hand-rolled `old.zip` unlink at its only call site go the same way, which removes a copy of the unlink-then-rename idiom rather than adding a fifth. Test 101 plants both kinds of leftover before the update pass: a stale file, which has to be clobbered so the `-M` abort can still restore the pass-1 copy, and a directory, which cannot be clobbered and has to be reported instead. Only the Windows leg arms the first half, since POSIX clobbers on its own; the second is armed everywhere. Test 37 gains a live update pass so the dead pass rotates onto an existing `old.zip`, which nothing covered. Two older bugs on the same path came out of the review and are filed separately: #774 (the `.bak` name collides with a mirrored file of the same name, reproduced) and #775 (a failed `filecreate()` on a chunked re-fetch commits the backup away). Closes #758 Signed-off-by: Xavier Roche <roche@httrack.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-27 11:10:17 +02:00
Xavier Roche	1ef3458b2e	An FTP re-fetch truncated the mirrored file before the transfer filecreate() emptied url_sav before a single byte had arrived, so a read error, a timeout, a short body or a local write error destroyed the previous copy. HTTP has taken a .bak aside since the #77 follow-up; FTP never did. Lift that backup out of back_wait()'s direct-to-disk path into back_refetch_backup() and call it from the FTP transfer too, so both inherit the same restore in back_finalize() rather than growing a second copy of the idiom. Closes #771 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Xavier Roche <roche@httrack.com>	2026-07-27 10:24:29 +02:00
Xavier Roche	56389103bf	An --update pass overwrites the previous WARC and leaves a page-less WACZ (#777 ) * warc: keep the previous archive when a pass has no bodies to replace it A second crawl into the same output reopened the WARC with "wb" and truncated it. On a cache-served pass nearly every URL comes back 304, so the new file held revisit records whose payloads had just been deleted, and the regenerated WACZ came out with zero page rows: a package that replays nothing, in place of one that replayed fine. The writer now builds into a sibling .tmp whenever an archive is already there, and only swaps it in at close if the result can stand on its own. A pass that only revisited URLs it did not re-download keeps the previous .warc.gz, .cdx and .wacz untouched and says so. What --update should ultimately mean for WARC output is still open; this only stops the silent destruction in the meantime. Closes #759 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Xavier Roche <roche@httrack.com> * warc: guard the segment swap and cover the rotated archive A run that lost a record or a segment must not replace a whole archive, and hts_rename_over unlinks its destination when the source is missing, so every segment has to be on disk before the first rename. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Xavier Roche <roche@httrack.com> * warc: force rotation in the segment test and guard its vacuity --warc-max-size 2000 never rotated under the harness's --robots=0, so the segment test was checking a single-file archive; a mutant that renamed only segment 0 passed it. 600 rotates, and --archive-min-files fails the test if a shrinking crawl ever stops producing the segments it checks. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Xavier Roche <roche@httrack.com> --------- Signed-off-by: Xavier Roche <roche@httrack.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-27 10:24:06 +02:00
Xavier Roche	a4db58f7b7	hts_finish_html_file is documented as skipping unchanged writes, and does not (#764 ) The MD5 comparison the comment promises left the tree in two steps: #467 extracted the function out of htsparse.c's HT_ADD_END macro without the skip branch, and #512 removed the //[HTML-MD5]// cache entry it read. Drop the parenthetical; the write is unconditional. Signed-off-by: Xavier Roche <roche@httrack.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-27 10:11:11 +02:00
Xavier Roche	1f9728d816	--purge-old deleted a file whose re-fetch never got a response (#765 ) * A failed re-fetch overwrote the mirrored file with the aborted read's debris A transfer that dies before a complete response has no body, yet the save path still consulted r.adr, which at that point holds whatever the aborted header read left behind: raw status-line bytes, or an empty buffer that truncated the file to zero on macOS. Require a successful transfer, as the empty-body half of the condition already did. Closes #748 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Xavier Roche <roche@httrack.com> * tests: assert the failed re-fetch keeps its bytes on every platform Test 93 filtered reset.bin out of its bucket lists because a connection killed before the status line surfaced differently per platform. It no longer does, so assert the resource like any other: unchanged, with the bytes pass 1 mirrored. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Xavier Roche <roche@httrack.com> * tests: make the re-fetch assertions catch a crawl that never re-fetched The new test passed with no second pass at all, so a regression that stopped re-fetching would have looked green. Assert the failure the fixture provokes, give stay.bin a fresh pass-2 body so a fix that stopped overwriting anything fails, and compare reset.bin by checksum rather than by length. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Xavier Roche <roche@httrack.com> * --purge-old deleted a file whose re-fetch never got a response A transfer that dies on the wire leaves the previously mirrored copy in place, but back_finalize() returned without noting it, so the URL fell out of new.lst and the end-of-update purge treated the file like a page that had vanished from the site. Note the surviving copy, the way the incomplete-transfer branch above already does, for the connection-level failures htsparse.c retries on. A deliberate skip (too big, MIME-excluded, cancelled) keeps its current fate. Closes #746 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Xavier Roche <roche@httrack.com> * Keep the copy for every failure, not just the five retryable codes A malformed status line, an oversized declared length or a mid-flight abort all land on STATUSCODE_INVALID, outside the retryable set, and the purge still ate the file. Invert the test: any failure keeps the copy except the codes that mean the engine passed the resource over on purpose. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Xavier Roche <roche@httrack.com> * tests: assert the retry exhaustion, not the per-platform failure message The message a cut connection produces depends on whether any bytes arrived, so matching it would fail on a runner that sees none. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Xavier Roche <roche@httrack.com> --------- Signed-off-by: Xavier Roche <roche@httrack.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-27 10:03:31 +02:00