Pin the naming contract: knobs and fixtures for content-independent naming

-#test=savename gains body= (leading body bytes via a temp url_sav
file) and cached= (a real one-entry cache, reopened read-only, whose
stored body is PNG magic); new rows and 01_zlib-savename-cached.test
pin that naming never depends on content or on the previously recorded
save name, only on headers. e2e fixtures (wrongtype.jpg served as
image/png, a gzip variant, a 16 KiB body, content that changes between
crawls) pin the wire-wins outcome across fresh and update passes. Any
future content-based tie-break must flip these rows explicitly.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Signed-off-by: Xavier Roche <roche@httrack.com>

src/htsback.c

@@ -2237,12 +2237,13 @@ int host_wait(httrackp * opt, lien_back * back) {
 
 static int slot_can_be_cleaned(const lien_back * back) {
   return (back->status == STATUS_READY) // ready
-    /* Check autoclean */
-    && (!back->testmode)        // not test mode
-    && (strnotempty(back->url_sav))     // filename exists
-    && (HTTP_IS_OK(back->r.statuscode)) // HTTP "OK"
-    && (back->r.size >= 0)      // size>=0
-    ;
+         /* Check autoclean */
+         && (!back->locked)   // not held by hts_wait_delayed (name pending)
+         && (!back->testmode) // not test mode
+         && (strnotempty(back->url_sav))     // filename exists
+         && (HTTP_IS_OK(back->r.statuscode)) // HTTP "OK"
+         && (back->r.size >= 0)              // size>=0
+      ;
 }
 
 static int slot_can_be_finalized(httrackp * opt, const lien_back * back) {
@@ -2891,10 +2892,10 @@ void back_wait(struct_back * sback, httrackp * opt, cache_back * cache,
           // range size hack old location
 
 #if HTS_DIRECTDISK
-          // Court-circuit:
-          // Peut-on stocker le fichier directement sur disque?
-          // Ahh que ca serait vachement mieux et que ahh que la mémoire vous dit merci!
-          if (back[i].status) {
+          // Shortcut: store the file directly on disk when possible,
+          // sparing memory
+          if (back[i].status &&
+              !back[i].locked) { // name still pending when locked
             if (back[i].r.is_write == 0) {      // mode mémoire
               if (back[i].r.adr == NULL) {      // rien n'a été écrit
                 if (!back[i].testmode) {        // pas mode test
@@ -3960,8 +3961,12 @@ void back_wait(struct_back * sback, httrackp * opt, cache_back * cache,
                               && (back[i].r.adr = (char *) malloct(2))) {
                             back[i].r.adr[0] = 0;
                           }
-                          hts_log_print(opt, LOG_TRACE, "finalizing empty");
-                          back_finalize(opt, cache, sback, i);
+                          /* locked = name pending; the waiter finalizes after
+                             patching url_sav (else: cached as .delayed, #5) */
+                          if (!back[i].locked) {
+                            hts_log_print(opt, LOG_TRACE, "finalizing empty");
+                            back_finalize(opt, cache, sback, i);
+                          }
                         } else if (!back[i].r.is_chunk) {       // pas de chunk
                           //if (back[i].r.http11!=2) {    // pas de chunk
                           back[i].is_chunk = 0;

src/htsname.c

@@ -138,37 +138,66 @@ static void cleanEndingSpaceOrDot(char *s) {
   }
 }
 
-/* Should the wire Content-Type override the URL's own extension when naming the
-   saved file? True when the type is patchable (may_unknown2) and either the URL
-   extension implies no specific type or the server declared a disagreeing one.
-   A URL extension mapping to a specific non-HTML type is kept only when the
-   server declared NO type (the HTS_UNKNOWN_MIME sentinel; the #267 mangle
-   guard): a typeless .png stays .png, but a .pdf explicitly served as text/html
-   is named .html. The sentinel rides the cache, so updates stay consistent. */
+/* Wire Content-Type vs URL extension: a patchable wire type wins over an
+   unspecific ext, the HTS_UNKNOWN_MIME sentinel keeps a specific non-HTML ext
+   (#267 guard), a declared disagreement is CONTESTED. Sentinel and verdict
+   ride the cache, so updates stay consistent. */
+typedef enum wire_verdict {
+  WIRE_KEEPS_EXT,
+  WIRE_WINS,
+  WIRE_CONTESTED
+} wire_verdict;
+
+static wire_verdict wire_ext_verdict(httrackp *opt, const char *wiremime,
+                                     const char *file, char *urlmime,
+                                     size_t urlmime_size) {
+  if (may_unknown2(opt, wiremime, file))
+    return WIRE_KEEPS_EXT; /* type kept verbatim (keep-list / bogus-multiple) */
+  urlmime[0] = '\0';
+  /* type implied by the URL extension, only when confidently known (flag 0) */
+  if (!get_httptype_sized(opt, urlmime, urlmime_size, file, 0))
+    return WIRE_WINS; /* URL ext implies no known type */
+  if (strfield2(wiremime, urlmime))
+    return WIRE_KEEPS_EXT; /* agreement (no .htm->.html churn) */
+  if (!is_hypertext_mime(opt, urlmime, file) &&
+      strfield2(wiremime, HTS_UNKNOWN_MIME))
+    return WIRE_KEEPS_EXT; /* no declared type */
+  return WIRE_CONTESTED;
+}
+
 static int wire_patches_ext(httrackp *opt, const char *wiremime,
                             const char *file) {
   char urlmime[256];
 
-  if (may_unknown2(opt, wiremime, file))
-    return 0; /* type kept verbatim (keep-list / bogus-multiple) */
-  urlmime[0] = '\0';
-  /* type implied by the URL extension, only when confidently known (flag 0) */
-  if (!get_httptype_sized(opt, urlmime, sizeof(urlmime), file, 0))
-    return 1; /* URL ext implies no known type: trust the wire type */
-  if (strfield2(wiremime, urlmime))
-    return 0; /* wire agrees with the ext: keep it (no .htm->.html churn) */
-  /* wire disagrees with a specific non-HTML URL ext. Keep the ext only when
-     the server declared no type (the sentinel); an explicitly declared type,
-     even text/html, is trusted, so a binary-looking URL that really serves
-     HTML (login/error interstitial, soft-404) is named .html. */
-  if (!is_hypertext_mime(opt, urlmime, file) &&
-      strfield2(wiremime, HTS_UNKNOWN_MIME))
+  switch (wire_ext_verdict(opt, wiremime, file, urlmime, sizeof(urlmime))) {
+  case WIRE_KEEPS_EXT:
     return 0;
+  case WIRE_WINS:
+    return 1;
+  case WIRE_CONTESTED:
+    break; /* no content evidence is consulted today: trust the wire */
+  }
   return 1;
 }
 
-// forme le nom du fichier à sauver (save) à partir de fil et adr
-// système intelligent, qui renomme en cas de besoin (exemple: deux INDEX.HTML et index.html)
+/* Wire-metadata name change: a Content-Disposition filename wins (returns 2),
+   else the declared type's ext when wire_patches_ext() allows (returns 1),
+   else 0. ext receives the new extension or replacement filename. */
+static int resolve_extension(httrackp *opt, const char *cdispo,
+                             const char *contenttype, const char *fil,
+                             char *ext, size_t ext_size) {
+  if (strnotempty(cdispo)) {
+    strlcpybuff(ext, cdispo, ext_size);
+    return 2;
+  }
+  if (wire_patches_ext(opt, contenttype, fil) &&
+      give_mimext(ext, ext_size, contenttype))
+    return 1;
+  return 0;
+}
+
+// Build the local save name (save) from adr/fil; renames on collision
+// (e.g. INDEX.HTML vs index.html).
 int url_savename(lien_adrfilsave *const afs,
                  lien_adrfil *const former,
                  const char *referer_adr, const char *referer_fil, 
@@ -405,45 +434,23 @@ int url_savename(lien_adrfilsave *const afs,
 
   // si option check_type activée
   if (is_html < 0 && opt->check_type && !ext_chg) {
-    int ishtest = 0;
-
     if (protocol != PROTOCOL_FILE
         && 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->savename_delayed == HTS_SAVENAME_DELAYED_HARD ||
-            (ishtest = ishtml(opt, fil)) <
-                0) { // unsure whether it's html or a file
+            ishtml(opt, fil) < 0) { // unsure whether it's html or a file
           // lire dans le cache
           htsblk r = cache_read_including_broken(opt, cache, adr, fil); // test uniquement
 
-          if (r.statuscode != -1) {     // pas d'erreur de lecture cache
-            char s[32];
-
-            s[0] = '\0';
+          if (r.statuscode != -1) { // cache entry read OK
             hts_log_print(opt, LOG_DEBUG, "Testing link type (from cache) %s%s",
                           adr_complete, fil_complete);
             if (!HTTP_IS_REDIRECT(r.statuscode)) {
-              if (strnotempty(r.cdispo)) {        /* filename given */
-                ext_chg = 2;      /* change filename */
-                strcpybuff(ext, r.cdispo);
-              } else if (wire_patches_ext(opt, r.contenttype, fil)) {
-                if (give_mimext(s, sizeof(s),
-                                r.contenttype)) { // recognized extension
-                  ext_chg = 1;
-                  strcpybuff(ext, s);
-                }
-              }
+              ext_chg = resolve_extension(opt, r.cdispo, r.contenttype, fil,
+                                          ext, sizeof(ext));
             }
-#ifdef DEFAULT_BIN_EXT
-            // no extension and potentially bogus
-            else if (ishtest == -2) {
-              ext_chg = 1;
-              strcpybuff(ext, DEFAULT_BIN_EXT + 1);
-            }
-#endif
-            //
           } else if (opt->savename_delayed != HTS_SAVENAME_DELAYED_HARD &&
                      is_userknowntype(opt, fil)) { /* PATCH BY BRIAN SCHRÖDER.
                               Lookup mimetype not only by extension,
@@ -467,22 +474,11 @@ int url_savename(lien_adrfilsave *const afs,
           // fail later
           else if (opt->savename_delayed != HTS_SAVENAME_DELAYED_NONE &&
                    !opt->state.stop) {
-            // Check if the file is ready in backing. We basically take the same logic as later.
-            // FIXME: we should cleanup and factorize this unholy mess
+            // Check if the file is ready in backing.
             if (headers != NULL && headers->status >= 0 && !is_redirect) {
-              if (strnotempty(headers->r.cdispo)) {        /* filename given */
-                ext_chg = 2;      /* change filename */
-                strcpybuff(ext, headers->r.cdispo);
-              } else if (wire_patches_ext(opt, headers->r.contenttype,
-                                          headers->url_fil)) {
-                char s[16];
-                if (give_mimext(
-                        s, sizeof(s),
-                        headers->r.contenttype)) { // recognized extension
-                  ext_chg = 1;
-                  strcpybuff(ext, s);
-                }
-              }
+              ext_chg = resolve_extension(opt, headers->r.cdispo,
+                                          headers->r.contenttype,
+                                          headers->url_fil, ext, sizeof(ext));
             }
             else if (mime_type != NULL) {
               ext[0] = '\0';
@@ -500,13 +496,6 @@ int url_savename(lien_adrfilsave *const afs,
                   if (!may_unknown2(opt, mime_type, fil)) {
                     ext_chg = 1;
                   }
-#ifdef DEFAULT_BIN_EXT
-                  // no extension and potentially bogus
-                  else if (ishtml(opt, fil) == -2) {
-                    ext_chg = 1;
-                    strcpybuff(ext, DEFAULT_BIN_EXT + 1);
-                  }
-#endif
                 } else {
                   ext_chg = 0;
                 }
@@ -696,30 +685,10 @@ int url_savename(lien_adrfilsave *const afs,
                     // libérer emplacement backing
                   }
 
-                  {             // pas d'erreur, changer type?
-                    char s[16];
-
-                    s[0] = '\0';
-                    if (strnotempty(back[b].r.cdispo)) {        /* filename given */
-                      ext_chg = 2;      /* change filename */
-                      strcpybuff(ext, back[b].r.cdispo);
-                    } else if (wire_patches_ext(opt, back[b].r.contenttype,
-                                                back[b].url_fil)) {
-                      if (give_mimext(
-                              s, sizeof(s),
-                              back[b].r.contenttype)) { // recognized extension
-                        ext_chg = 1;
-                        strcpybuff(ext, s);
-                      }
-                    }
-#ifdef DEFAULT_BIN_EXT
-                    // no extension and potentially bogus
-                    else if (ishtest == -2) {
-                      ext_chg = 1;
-                      strcpybuff(ext, DEFAULT_BIN_EXT + 1);
-                    }
-#endif
-                  }
+                  // no error: change the type?
+                  ext_chg = resolve_extension(
+                      opt, back[b].r.cdispo, back[b].r.contenttype,
+                      back[b].url_fil, ext, sizeof(ext));
                 }
                 // FIN Si non déplacé, forcer type?
 

src/htsparse.c

@@ -4845,6 +4845,9 @@ int hts_wait_delayed(htsmoduleStruct * str, lien_adrfilsave *afs,
 
           /* Still have a back reference */
           if (b >= 0) {
+            /* Patch destination filename for direct-to-disk mode, BEFORE any
+               finalize: it records and caches the entry under url_sav */
+            strcpybuff(back[b].url_sav, afs->save);
             /* Finalize now as we have the type */
             if (back[b].status == STATUS_READY) {
               if (!back[b].finalized) {
@@ -4852,8 +4855,6 @@ int hts_wait_delayed(htsmoduleStruct * str, lien_adrfilsave *afs,
                 back_finalize(opt, cache, sback, b);
               }
             }
-            /* Patch destination filename for direct-to-disk mode */
-            strcpybuff(back[b].url_sav, afs->save);
           }
 
         }                       // b >= 0

src/htsselftest.c

@@ -1093,35 +1093,202 @@ static int st_resolve(httrackp *opt, int argc, char **argv) {
   return 0;
 }
 
+/* Extra args are key=value: adr= cdispo= statuscode= status= strip= urlhack=
+   no-www= no-slash= no-query= n83= type=, plus repeatable prior=adr|fil|sav
+   registering an already-crawled link (dedup/collision paths). */
+/* Parse raw response-header lines and print the naming-relevant fields. */
+static int st_header(httrackp *opt, int argc, char **argv) {
+  htsblk r;
+  int i;
+
+  (void) opt;
+  if (argc < 1) {
+    fprintf(stderr, "header: needs at least one raw header line\n");
+    return 1;
+  }
+  memset(&r, 0, sizeof(r));
+  for (i = 0; i < argc; i++) {
+    char BIGSTK line[HTS_URLMAXSIZE * 2];
+
+    strcpybuff(line, argv[i]);
+    treathead(NULL, "www.example.com", "/", &r, line);
+  }
+  printf("contenttype=%s cdispo=%s\n", r.contenttype, r.cdispo);
+  return 0;
+}
+
+/* Decode a body argument ("hex:FFD8.." or literal text) into buf. */
+static size_t st_decode_body(const char *arg, char *buf, size_t size) {
+  size_t n = 0;
+
+  if (strncmp(arg, "hex:", 4) == 0) {
+    const char *s = arg + 4;
+
+    for (; s[0] != '\0' && s[1] != '\0' && n + 1 < size; s += 2) {
+      unsigned int byte;
+
+      if (sscanf(s, "%2x", &byte) != 1)
+        break;
+      buf[n++] = (char) byte;
+    }
+  } else {
+    n = strlen(arg);
+    if (n >= size)
+      n = size - 1;
+    memcpy(buf, arg, n);
+  }
+  buf[n] = '\0';
+  return n;
+}
+
 static int st_savename(httrackp *opt, int argc, char **argv) {
   lien_adrfilsave afs;
   cache_back cache;
   struct_back *sback;
   hash_struct hash;
   lien_back headers;
+  const char *adr = "www.example.com";
+  const char *cdispo = NULL;
+  const char *body = NULL;
+  const char *cached = NULL;
+  const char *bodyfile = "st-savename-body.tmp";
+  int statuscode = HTTP_OK, status = 0;
+  int i;
 
   if (argc < 2) {
     fprintf(stderr, "savename: needs a fil and a content-type\n");
     return 1;
   }
+  /* knobs first: hash_init and the prior links depend on them */
+  for (i = 2; i < argc; i++) {
+    const char *const a = argv[i];
+
+    if (strncmp(a, "adr=", 4) == 0)
+      adr = a + 4;
+    else if (strncmp(a, "cdispo=", 7) == 0)
+      cdispo = a + 7;
+    else if (strncmp(a, "statuscode=", 11) == 0)
+      statuscode = atoi(a + 11);
+    else if (strncmp(a, "status=", 7) == 0)
+      status = atoi(a + 7);
+    else if (strncmp(a, "strip=", 6) == 0)
+      StringCopy(opt->strip_query, a + 6);
+    else if (strncmp(a, "urlhack=", 8) == 0)
+      opt->urlhack = atoi(a + 8) ? HTS_TRUE : HTS_FALSE;
+    else if (strncmp(a, "no-www=", 7) == 0)
+      opt->no_www_dedup = atoi(a + 7) ? HTS_TRUE : HTS_FALSE;
+    else if (strncmp(a, "no-slash=", 9) == 0)
+      opt->no_slash_dedup = atoi(a + 9) ? HTS_TRUE : HTS_FALSE;
+    else if (strncmp(a, "no-query=", 9) == 0)
+      opt->no_query_dedup = atoi(a + 9) ? HTS_TRUE : HTS_FALSE;
+    else if (strncmp(a, "n83=", 4) == 0)
+      opt->savename_83 = atoi(a + 4);
+    else if (strncmp(a, "type=", 5) == 0)
+      opt->savename_type = atoi(a + 5);
+    else if (strncmp(a, "body=", 5) == 0)
+      body = a + 5;
+    else if (strncmp(a, "cached=", 7) == 0)
+      cached = a + 7;
+    else if (strncmp(a, "prior=", 6) != 0) {
+      fprintf(stderr, "savename: unknown arg '%s'\n", a);
+      return 1;
+    }
+  }
   memset(&afs, 0, sizeof(afs));
-  strcpybuff(afs.af.adr, "www.example.com");
+  strcpybuff(afs.af.adr, adr);
   strcpybuff(afs.af.fil, argv[0]);
 
   memset(&cache, 0, sizeof(cache));
-  cache.hashtable = (void *) coucal_new(0);
+  if (cached != NULL) { /* cached=<content-type>|<save name> */
+    char *dup = strdupt(cached);
+    char *const sep = strchr(dup, '|');
+    char locbuf[64] = "";
+    htsblk cr;
+
+    if (sep == NULL) {
+      fprintf(stderr, "savename: cached needs ctype|save\n");
+      return 1;
+    }
+    *sep = '\0';
+    /* one-entry cache in cwd, reopened read-only; body is PNG magic on
+       purpose: naming must not depend on stored content */
+    StringCopy(opt->path_log, "");
+    cache.type = 1;
+    cache.log = cache.errlog = stderr;
+    cache.hashtable = coucal_new(0);
+    cache_init(&cache, opt);
+    hts_init_htsblk(&cr);
+    cr.statuscode = HTTP_OK;
+    strcpybuff(cr.msg, "OK");
+    strcpybuff(cr.contenttype, dup);
+    cr.location = locbuf;
+    cr.adr = strdupt("\x89PNG\r\n\x1a\n");
+    cr.size = 8;
+    cache_add(opt, &cache, &cr, adr, argv[0], sep + 1, 1, NULL);
+    freet(cr.adr);
+    if (cache.zipOutput != NULL) {
+      zipClose(cache.zipOutput, NULL);
+      cache.zipOutput = NULL;
+    }
+    memset(&cache, 0, sizeof(cache));
+    cache.type = 1;
+    cache.log = cache.errlog = stderr;
+    cache.hashtable = coucal_new(0);
+    cache.ro = 1;
+    cache_init(&cache, opt);
+    freet(dup);
+  } else {
+    cache.hashtable = (void *) coucal_new(0);
+  }
 
   sback = back_new(opt, opt->maxsoc * 32 + 1024);
+  /* same wiring as hts_mirror (htscore.c) */
   hash_init(opt, &hash, opt->urlhack);
+  hash.liens = (const lien_url *const *const *) &opt->liens;
+  opt->hash = &hash;
+  hts_record_init(opt);
+
+  for (i = 2; i < argc; i++) {
+    if (strncmp(argv[i], "prior=", 6) == 0) {
+      char *dup = strdupt(argv[i] + 6);
+      char *const p1 = strchr(dup, '|');
+      char *const p2 = p1 != NULL ? strchr(p1 + 1, '|') : NULL;
+
+      if (p2 == NULL) {
+        fprintf(stderr, "savename: prior needs adr|fil|sav\n");
+        return 1;
+      }
+      *p1 = *p2 = '\0';
+      if (!hts_record_link(opt, dup, p1 + 1, p2 + 1, "", "", NULL))
+        return 1;
+      freet(dup);
+    }
+  }
 
   memset(&headers, 0, sizeof(headers));
-  headers.status = 0;
-  headers.r.statuscode = HTTP_OK;
+  headers.status = status;
+  headers.r.statuscode = statuscode;
   strcpybuff(headers.r.contenttype, argv[1]);
+  if (cdispo != NULL)
+    strcpybuff(headers.r.cdispo, cdispo);
   strcpybuff(headers.url_fil, argv[0]);
+  if (body != NULL) { /* leading body bytes, exposed via url_sav */
+    char BIGSTK data[1024];
+    const size_t n = st_decode_body(body, data, sizeof(data));
+    FILE *const fp = fopen(bodyfile, "wb");
+
+    if (fp == NULL || fwrite(data, 1, n, fp) != n) {
+      fprintf(stderr, "savename: can not write %s\n", bodyfile);
+      return 1;
+    }
+    fclose(fp);
+    strcpybuff(headers.url_sav, bodyfile);
+  }
 
   url_savename(&afs, NULL, NULL, NULL, opt, sback, &cache, &hash, 0, 0,
                &headers);
+  if (body != NULL)
+    (void) UNLINK(bodyfile);
   printf("savename: %s\n", afs.save);
   return 0;
 }
@@ -1924,8 +2091,10 @@ static const struct selftest_entry {
      st_relative},
     {"resolve", "<link> <adr> <fil>", "resolve a link against an origin",
      st_resolve},
-    {"savename", "<fil> <content-type>", "local save-name for a URL",
-     st_savename},
+    {"header", "<raw-header-line> ...", "response header-line parsing",
+     st_header},
+    {"savename", "<fil> <content-type> [key=value ...]",
+     "local save-name for a URL", st_savename},
     {"cache", "<dir>", "cache read/write round-trip self-test", st_cache},
     {"cache-golden", "<dir> [regen]", "frozen cache-format read self-test",
      st_cache_golden},

tests/01_engine-header.test

@@ -0,0 +1,29 @@
+#!/bin/bash
+#
+
+set -euo pipefail
+
+# Response header-line parsing (treathead via -#test=header <raw-line> ...).
+# Isolates the wire layer from url_savename, which strips traversal on its own.
+
+hdr() {
+    local want="$1"
+    shift
+    out="$(httrack -O /dev/null -#test=header "$@" | grep '^contenttype=')"
+    test "$out" == "$want" || {
+        echo "FAIL: $* -> '$out' (want '$want')"
+        exit 1
+    }
+}
+
+hdr 'contenttype=application/pdf cdispo=' 'Content-Type: application/pdf'
+
+# filename= is honored quoted or bare.
+hdr 'contenttype= cdispo=report.pdf' \
+    'Content-Disposition: attachment; filename="report.pdf"'
+hdr 'contenttype= cdispo=report.pdf' \
+    'Content-Disposition: attachment; filename=report.pdf'
+
+# Path components in the filename are dropped on the wire (RFC 2616).
+hdr 'contenttype= cdispo=evil.pdf' \
+    'Content-Disposition: attachment; filename="../../evil.pdf"'

tests/01_engine-savename.test

@@ -3,13 +3,38 @@
 
 set -euo pipefail
 
-# Local save-name extension resolution (url_savename via -#test=savename <fil> <content-type>).
-# Asserts on the basename of "savename: <path>".
+# Local save-name resolution (url_savename via -#test=savename <fil> <content-type> [key=value ...]).
+# name() asserts on the basename, full() on the whole path; prior= registers an
+# already-crawled link whose sav is rooted under the -O path (/dev/null here).
+
+# resolve httrack before cd: make check puts a RELATIVE ../src on PATH
+httrack_bin=$(cd "$(dirname "$(command -v httrack)")" && pwd)/httrack
+
+# scratch dir: body= and cached= write temp files (st-savename-body.tmp, hts-cache/)
+scratch=$(mktemp -d)
+trap 'rm -rf "$scratch"' EXIT
+cd "$scratch"
+
+run() {
+    "$httrack_bin" -O /dev/null -#test=savename "$@" | sed -n 's/^savename: //p'
+}
 
 name() {
-    out="$(httrack -O /dev/null -#test=savename "$1" "$2" | sed -n 's/^savename: //p')"
-    test "${out##*/}" == "$3" || {
-        echo "FAIL: '$1' '$2' -> '$out' (want '$3')"
+    local fil="$1" ctype="$2" want="$3"
+    shift 3
+    out="$(run "$fil" "$ctype" "$@")"
+    test "${out##*/}" == "$want" || {
+        echo "FAIL: '$fil' '$ctype' $* -> '$out' (want '$want')"
+        exit 1
+    }
+}
+
+full() {
+    local fil="$1" ctype="$2" want="$3"
+    shift 3
+    out="$(run "$fil" "$ctype" "$@")"
+    test "$out" == "$want" || {
+        echo "FAIL: '$fil' '$ctype' $* -> '$out' (want '$want')"
         exit 1
     }
 }
@@ -39,3 +64,96 @@ name '/types/data.json' 'application/json' 'data.json'
 
 # Agreeing type must not rewrite the extension's casing (no strip-and-reappend).
 name '/x.JPG' 'image/jpeg' 'x.JPG'
+
+# A Content-Disposition filename replaces the URL name outright.
+name '/x.php' 'application/pdf' 'report.pdf' cdispo=report.pdf
+name '/download' 'text/html' 'setup.exe' cdispo=setup.exe
+
+# Reserved characters in a hostile Content-Disposition name are sanitized.
+name '/x.php' 'application/pdf' 'set_up.exe' 'cdispo=set:up.exe'
+
+# The md5-of-query suffix lands inside a Content-Disposition name too.
+name '/x.php?id=1' 'application/pdf' 'report681a.pdf' cdispo=report.pdf
+
+# Still-downloading path (status=-1): mime drives the ext, cdispo is ignored
+# there (the deliberately unfolded 4th resolve_extension variant).
+name '/x.pdf' 'text/html' 'x.html' status=-1
+name '/x.html' 'text/html' 'x.html' status=-1
+name '/x.php' 'application/pdf' 'x.pdf' status=-1 cdispo=report.pdf
+
+# Contested type (wire disagrees with a specific ext): the wire is trusted and
+# body bytes are not consulted; pinned so a content-based tie-break shows up
+# as an explicit flip of these rows.
+name '/photo.jpg' 'image/png' 'photo.png' body=hex:FFD8FFE000104A46
+name '/photo.jpg' 'image/png' 'photo.png' body=hex:89504E470D0A1A0A
+name '/photo.jpg' 'image/png' 'photo.png'
+name '/doc.pdf' 'text/html' 'doc.html' body=hex:255044462D312E34
+name '/doc.pdf' 'text/html' 'doc.html' 'body=<html><body>soft 404</body></html>'
+name '/style.css' 'image/png' 'style.png' 'body=body { }'
+
+# A redirect answer resolves nothing: delayed placeholder name.
+name '/x.php' 'text/html' 'x.0.delayed' statuscode=301
+
+# Root and query-only URLs get index + the md5-of-query suffix.
+name '/' 'text/html' 'index.html'
+name '/?a=1' 'text/html' 'index3872.html'
+
+# Same URL crawled before: reuse its sav verbatim (case preserved).
+full '/X.PHP' 'text/html' 'www.example.com/CASE.HTML' \
+    'prior=www.example.com|/X.PHP|www.example.com/CASE.HTML'
+
+# Another URL owns the name: collision suffix -2, then -3, case-insensitively.
+name '/x.php' 'text/html' 'x-2.html' \
+    'prior=www.example.com|/other.html|/dev/null/www.example.com/x.html'
+name '/x.php' 'text/html' 'x-3.html' \
+    'prior=www.example.com|/o1.html|/dev/null/www.example.com/x.html' \
+    'prior=www.example.com|/o2.html|/dev/null/www.example.com/x-2.html'
+name '/INDEX.HTML' 'text/html' 'INDEX-2.HTML' \
+    'prior=www.example.com|/index.html|/dev/null/www.example.com/index.html'
+
+# Same basename in another directory is NOT a collision.
+name '/x.php' 'text/html' 'x.html' \
+    'prior=www.example.com|/sub/x.html|/dev/null/www.example.com/sub/x.html'
+
+# 8-3 modes: DOS truncates every component to 8+3, ISO9660 level 2 to 31.
+full '/directory-long/verylongfilename.html' 'text/html' \
+    '/dev/null/EXAMPLE/DIRECTOR/VERYLONG.HTM' n83=1
+full '/directory-long/verylongfilename.html' 'text/html' \
+    '/dev/null/EXAMPLE_C/DIRECTORY_LONG/VERYLONGFILENAME.HTM' n83=2
+name '/verylongfilename.php' 'text/html' 'VERYLO-2.HTM' n83=1 \
+    'prior=www.example.com|/other.html|/dev/null/EXAMPLE/VERYLONG.HTM'
+
+# urlhack dedup (#271): // collapse and www-strip map to the prior link's sav;
+# the per-feature negatives opt out and take a fresh name.
+full '/a//b.php' 'text/html' '/dev/null/www.example.com/a/PRIOR.html' \
+    'prior=www.example.com|/a/b.php|/dev/null/www.example.com/a/PRIOR.html'
+full '/a//b.php' 'text/html' '/dev/null/www.example.com/a/b.html' no-slash=1 \
+    'prior=www.example.com|/a/b.php|/dev/null/www.example.com/a/PRIOR.html'
+full '/w.php' 'text/html' '/dev/null/www.example.com/W-PRIOR.html' adr=example.com \
+    'prior=www.example.com|/w.php|/dev/null/www.example.com/W-PRIOR.html'
+full '/w.php' 'text/html' '/dev/null/example.com/w.html' adr=example.com no-www=1 \
+    'prior=www.example.com|/w.php|/dev/null/www.example.com/W-PRIOR.html'
+
+# Distinct URLs must stay distinct under urlhack (no over-normalization).
+full '/a//b.php' 'text/html' '/dev/null/www.example.com/a/b.html' \
+    'prior=www.example.com|/a/c.php|/dev/null/www.example.com/a/C-PRIOR.html'
+
+# --strip-query (#112): stripped key dedups onto the prior sav; without the
+# option the same URLs stay distinct.
+full '/page.php?id=3&sid=42' 'text/html' '/dev/null/www.example.com/PAGE-PRIOR.html' \
+    strip=sid 'prior=www.example.com|/page.php?id=3|/dev/null/www.example.com/PAGE-PRIOR.html'
+full '/page.php?id=3&sid=42' 'text/html' '/dev/null/www.example.com/page475b.html' \
+    'prior=www.example.com|/page.php?id=3|/dev/null/www.example.com/PAGE-PRIOR.html'
+
+# A kept key that differs must still block the dedup (no over-stripping).
+full '/page.php?id=3&sid=42' 'text/html' '/dev/null/www.example.com/page475b.html' \
+    strip=sid 'prior=www.example.com|/page.php?id=4|/dev/null/www.example.com/PAGE-PRIOR.html'
+
+# Hostile fils stay rooted under the mirror: ../ (raw or %2e-encoded) drops out,
+# control characters become spaces, oversized names cap at 210 chars (the cap
+# can chop the extension off entirely).
+full '/../../etc/passwd' 'text/html' '/dev/null/www.example.com///etc/passwd.html'
+full '/%2e%2e/%2e%2e/etc/passwd' 'text/html' '/dev/null/www.example.com///etc/passwd.html'
+full '/x.php' 'application/pdf' '/dev/null/www.example.com///evil.exe' 'cdispo=../../evil.exe'
+name $'/evil\rname\t.php' 'text/html' 'evil name .html'
+name "/$(printf 'a%.0s' {1..300}).php" 'text/html' "$(printf 'a%.0s' {1..210})"

tests/01_zlib-savename-cached.test

@@ -0,0 +1,33 @@
+#!/bin/bash
+#
+
+set -euo pipefail
+
+# Update-run naming from a real cache entry (-#test=savename cached=<ctype>|<save>).
+# Named 01_zlib-*: the cache writer needs zlib, which the MSan job can't run.
+
+# resolve httrack before cd: make check puts a RELATIVE ../src on PATH
+httrack_bin=$(cd "$(dirname "$(command -v httrack)")" && pwd)/httrack
+
+scratch=$(mktemp -d)
+trap 'rm -rf "$scratch"' EXIT
+cd "$scratch"
+
+name() {
+    local fil="$1" ctype="$2" want="$3"
+    shift 3
+    out="$("$httrack_bin" -O /dev/null -#test=savename "$fil" "$ctype" "$@" | sed -n 's/^savename: //p')"
+    test "${out##*/}" == "$want" || {
+        echo "FAIL: '$fil' '$ctype' $* -> '$out' (want '$want')"
+        exit 1
+    }
+}
+
+# Names are re-derived from the stored headers on every run: neither the
+# recorded save name nor the cached body bytes change the verdict (pinned).
+name '/photo.jpg' 'image/png' 'photo.png' 'cached=image/png|www.example.com/photo.jpg'
+name '/photo.jpg' 'image/png' 'photo.png' 'cached=image/png|www.example.com/photo.png'
+name '/photo.jpg' 'image/jpeg' 'photo.jpg' 'cached=image/jpeg|www.example.com/photo.png'
+name '/style.css' 'image/png' 'style.png' 'cached=image/png|www.example.com/style.css'
+# agreement keeps the URL ext verbatim (.jpeg), never canonicalized to .jpg
+name '/photo.jpeg' 'image/jpeg' 'photo.jpeg' 'cached=image/jpeg|www.example.com/photo.jpeg'

tests/15_local-types.test

@@ -15,6 +15,10 @@ bash "$top_srcdir/tests/local-crawl.sh" --errors 0 \
     --found 'types/photo.png' \
     --found 'types/doc.pdf' \
     --found 'types/lie.html' --not-found 'types/lie.png' \
+    --found 'types/wrongtype.png' --not-found 'types/wrongtype.jpg' \
+    --found 'types/bigtype.png' --not-found 'types/bigtype.jpg' \
+    --found 'types/packed.png' --not-found 'types/packed.jpg' \
+    --found 'types/mutant.png' --not-found 'types/mutant.jpg' \
     --found 'types/report.html' --not-found 'types/report.pdf' \
     --found 'types/page.htm' --not-found 'types/page.html' \
     --found 'types/script.js' \

tests/18_local-update.test

@@ -12,4 +12,7 @@ bash "$top_srcdir/tests/local-crawl.sh" --errors 0 --rerun \
     --found 'types/report.html' --not-found 'types/report.pdf' \
     --found 'types/notype.png' --not-found 'types/notype.html' \
     --found 'types/lie.html' \
+    --found 'types/wrongtype.png' --not-found 'types/wrongtype.jpg' \
+    --found 'types/packed.png' --not-found 'types/packed.jpg' \
+    --found 'types/mutant.png' --not-found 'types/mutant.jpg' \
     httrack 'BASEURL/types/index.html'

tests/32_local-cdispo.test

@@ -0,0 +1,17 @@
+#!/bin/bash
+#
+# Content-Disposition names the saved file: the attachment filename replaces
+# the URL-derived name, and a traversal filename is reduced to its last
+# component, inside the mirror.
+
+set -euo pipefail
+
+: "${top_srcdir:=..}"
+
+bash "$top_srcdir/tests/local-crawl.sh" --errors 0 \
+    --found 'cdispo/report.pdf' \
+    --file-matches 'cdispo/report.pdf' '%PDF' \
+    --not-found 'cdispo/fetch.pdf' \
+    --found 'cdispo/evil.pdf' \
+    --not-found 'evil.pdf' \
+    httrack 'BASEURL/cdispo/index.html'

tests/33_local-delayed.test

@@ -0,0 +1,20 @@
+#!/bin/bash
+#
+# Degenerate delayed-type paths (#5/#107 family): redirects that never resolve
+# a name must drop cleanly -- no .delayed leftovers (audited by local-crawl.sh),
+# no "bogus state" cache warnings, resolvable links still land correctly.
+
+set -euo pipefail
+
+: "${top_srcdir:=..}"
+
+bash "$top_srcdir/tests/local-crawl.sh" --rerun --errors 0 \
+    --found 'delayed/real.pdf' \
+    --file-matches 'delayed/real.pdf' '%PDF' \
+    --found 'delayed/notype.bin.html' \
+    --found 'delayed/empty.html' \
+    --not-found 'delayed/noloc.html' \
+    --not-found 'delayed/selfloop.html' \
+    --not-found 'delayed/chain9.pdf' \
+    --log-not-found 'bogus state' \
+    httrack 'BASEURL/delayed/index.html'

tests/Makefile.am

@@ -38,6 +38,7 @@ TESTS = \
 	01_engine-ftp-line.test \
 	01_engine-ftp-userpass.test \
 	01_engine-hashtable.test \
+	01_engine-header.test \
 	01_engine-idna.test \
 	01_engine-escape-room.test \
 	01_engine-inplace-escape.test \
@@ -63,6 +64,7 @@ TESTS = \
 	01_zlib-cache.test \
 	01_zlib-cache-golden.test \
 	01_zlib-cache-writefail.test \
+	01_zlib-savename-cached.test \
 	02_manpage-regen.test \
 	02_update-cache.test \
 	10_crawl-simple.test \
@@ -91,6 +93,8 @@ TESTS = \
 	28_local-pause.test \
 	29_local-redirect-fragment.test \
 	30_local-fragment-link.test \
-	31_local-javaclass.test
+	31_local-javaclass.test \
+	32_local-cdispo.test \
+	33_local-delayed.test
 
 CLEANFILES = check-network_sh.cache

tests/local-crawl.sh

@@ -246,6 +246,14 @@ done
 test -n "$hostroot" || die "could not find host root under $out"
 debug "host root: $hostroot"
 
+# A completed crawl must leave no .delayed temporaries (issue #107)
+info "checking for leftover .delayed files"
+leftovers=$(find "$out" -name '*.delayed' 2>/dev/null | head -5)
+if test -z "$leftovers"; then result "OK"; else
+    result "leftover: $leftovers"
+    exit 1
+fi
+
 # --- audit -------------------------------------------------------------------
 i=0
 while test "$i" -lt "${#audit[@]}"; do

tests/local-server.py

@@ -14,6 +14,7 @@ stdlib only (http.server + ssl) -- no new build or runtime dependency.
 """
 
 import argparse
+import gzip
 import os
 import time
 from http.server import SimpleHTTPRequestHandler, ThreadingHTTPServer
@@ -134,12 +135,14 @@ class Handler(SimpleHTTPRequestHandler):
 
     # --- type/extension matrix (issue #267 family) -------------------------
 
-    def send_raw(self, body, content_type):
+    def send_raw(self, body, content_type, extra_headers=()):
         """Send a raw body with an explicit Content-Type, or none at all when
         content_type is None (to observe httrack's typeless-file naming)."""
         self.send_response(200)
         if content_type is not None:
             self.send_header("Content-Type", content_type)
+        for name, value in extra_headers:
+            self.send_header(name, value)
         self.send_header("Content-Length", str(len(body)))
         self.end_headers()
         if self.command != "HEAD":
@@ -148,6 +151,8 @@ class Handler(SimpleHTTPRequestHandler):
     # Fake-binary blobs for the image/pdf/typeless cases.
     FAKE_PNG = b"\x89PNG\r\n\x1a\n" + b"\x00" * 64
     FAKE_PDF = b"%PDF-1.4\n" + b"\x00" * 64
+    FAKE_JPEG = b"\xff\xd8\xff\xe0" + b"\x00" * 64
+    BIG_JPEG = b"\xff\xd8\xff\xe0" + bytes(range(256)) * 64  # > sniff window
 
     # path -> (body, content_type); None sends no header, "" sends an empty
     # Content-Type value (no usable type, must be treated like None).
@@ -159,6 +164,8 @@ class Handler(SimpleHTTPRequestHandler):
         "/types/notype.pdf": (FAKE_PDF, None),
         "/types/emptyct.png": (FAKE_PNG, ""),
         "/types/lie.png": (FAKE_PNG, "text/html"),
+        "/types/wrongtype.jpg": (FAKE_JPEG, "image/png"),
+        "/types/bigtype.jpg": (BIG_JPEG, "image/png"),
         "/types/report.pdf": (b"<html><body>real page</body></html>", "text/html"),
         "/types/page.htm": (b"<html><body>htm page</body></html>", "text/html"),
         "/types/script.js": (b"var x = 1;\n", "application/javascript"),
@@ -176,6 +183,10 @@ class Handler(SimpleHTTPRequestHandler):
             '\t<a href="notype.pdf">notypepdf</a>\n'
             '\t<img src="emptyct.png" />\n'
             '\t<img src="lie.png" />\n'
+            '\t<img src="wrongtype.jpg" />\n'
+            '\t<img src="bigtype.jpg" />\n'
+            '\t<img src="mutant.jpg" />\n'
+            '\t<img src="packed.jpg" />\n'
             '\t<a href="report.pdf">report</a>\n'
             '\t<a href="page.htm">htm</a>\n'
             '\t<script src="script.js"></script>\n'
@@ -190,6 +201,25 @@ class Handler(SimpleHTTPRequestHandler):
         body, ctype = self.TYPE_MATRIX[path]
         self.send_raw(body, ctype)
 
+    # content changes between crawls: run 1 sniffs JPEG, the update pass must
+    # keep the run-1 name (recorded verdict) even though the body is now PNG
+    MUTANT_SEEN = set()
+
+    def route_types_mutant(self):
+        path = urlsplit(self.path).path
+        body = self.FAKE_PNG if path in self.MUTANT_SEEN else self.FAKE_JPEG
+        if self.command != "HEAD":
+            self.MUTANT_SEEN.add(path)
+        self.send_raw(body, "image/png")
+
+    # gzip on the wire: the sniff must see the decoded body, not the stream
+    def route_types_packed(self):
+        self.send_raw(
+            gzip.compress(self.FAKE_JPEG),
+            "image/png",
+            extra_headers=[("Content-Encoding", "gzip")],
+        )
+
     # --- MIME-type exclusion abort (issue #58) -----------------------------
     # A -mime:application/pdf filter must abort the transfer once the header
     # arrives, not download the whole body and discard it.
@@ -354,6 +384,27 @@ class Handler(SimpleHTTPRequestHandler):
         if self.command != "HEAD":
             self.wfile.write(body)
 
+    # Content-Disposition naming: the attachment filename replaces the
+    # URL-derived name; path components in it are stripped (RFC 2616).
+    CDISPO_NAMES = {
+        "/cdispo/fetch.php": "report.pdf",
+        "/cdispo/evil.php": "../../evil.pdf",
+    }
+
+    def route_cdispo_index(self):
+        self.send_html(
+            '\t<a href="fetch.php">report</a>\n' '\t<a href="evil.php">evil</a>\n'
+        )
+
+    def route_cdispo(self):
+        filename = self.CDISPO_NAMES[urlsplit(self.path).path]
+        cdispo = 'attachment; filename="%s"' % filename
+        self.send_raw(
+            self.FAKE_PDF,
+            "application/pdf",
+            extra_headers=[("Content-Disposition", cdispo)],
+        )
+
     # 302 whose Location carries a #fragment (#204): the fragment is a UA anchor
     # that must be dropped before the target is fetched. A leaked '#' reaches the
     # strict-server guard below and 400s.
@@ -369,6 +420,50 @@ class Handler(SimpleHTTPRequestHandler):
     def route_redir_target(self):
         self.send_raw(b"<html><body>redirect target</body></html>\n", "text/html")
 
+    # --- delayed-type degenerate paths (issues #5/#107) --------------------
+    def route_delayed_index(self):
+        self.send_html(
+            '\t<a href="noloc.php">noloc</a>\n'
+            '\t<a href="selfloop.php">selfloop</a>\n'
+            '\t<a href="chain1.php">chain</a>\n'
+            '\t<a href="redir.php">redir</a>\n'
+            '\t<a href="notype.bin">notype</a>\n'
+            '\t<a href="empty.php">empty</a>\n'
+        )
+
+    def send_redirect(self, location):
+        self.send_response(302, "Found")
+        if location is not None:
+            self.send_header("Location", location)
+        self.send_header("Content-Length", "0")
+        self.end_headers()
+
+    def route_delayed_noloc(self):
+        self.send_redirect(None)  # 302 without Location: name never resolves
+
+    def route_delayed_selfloop(self):
+        self.send_redirect("selfloop.php")
+
+    def route_delayed_chain(self):
+        # chain1..chain9: one more hop than the type-check redirect budget
+        n = int(urlsplit(self.path).path.rsplit("chain", 1)[1].split(".")[0])
+        if n < 9:
+            self.send_redirect("chain%d.php" % (n + 1))
+        else:
+            self.send_raw(self.FAKE_PDF, "application/pdf")
+
+    def route_delayed_redir(self):
+        self.send_redirect("real.pdf")
+
+    def route_delayed_realpdf(self):
+        self.send_raw(self.FAKE_PDF, "application/pdf")
+
+    def route_delayed_notype(self):
+        self.send_raw(self.FAKE_PDF, None)
+
+    def route_delayed_empty(self):
+        self.send_raw(b"", "text/html")  # 200 + Content-Length: 0
+
     ROUTES = {
         "/cookies/entrance.php": route_entrance,
         "/cookies/second.php": route_second,
@@ -384,6 +479,10 @@ class Handler(SimpleHTTPRequestHandler):
         "/types/notype.pdf": route_types,
         "/types/emptyct.png": route_types,
         "/types/lie.png": route_types,
+        "/types/wrongtype.jpg": route_types,
+        "/types/bigtype.jpg": route_types,
+        "/types/mutant.jpg": route_types_mutant,
+        "/types/packed.jpg": route_types_packed,
         "/types/report.pdf": route_types,
         "/types/page.htm": route_types,
         "/types/script.js": route_types,
@@ -406,6 +505,25 @@ class Handler(SimpleHTTPRequestHandler):
         "/mimex/index.html": route_mimex_index,
         "/mimex/blob.pdf": route_mimex_blob,
         "/mimex/real.html": route_mimex_real,
+        "/cdispo/index.html": route_cdispo_index,
+        "/cdispo/fetch.php": route_cdispo,
+        "/cdispo/evil.php": route_cdispo,
+        "/delayed/index.html": route_delayed_index,
+        "/delayed/noloc.php": route_delayed_noloc,
+        "/delayed/selfloop.php": route_delayed_selfloop,
+        "/delayed/redir.php": route_delayed_redir,
+        "/delayed/real.pdf": route_delayed_realpdf,
+        "/delayed/notype.bin": route_delayed_notype,
+        "/delayed/empty.php": route_delayed_empty,
+        "/delayed/chain1.php": route_delayed_chain,
+        "/delayed/chain2.php": route_delayed_chain,
+        "/delayed/chain3.php": route_delayed_chain,
+        "/delayed/chain4.php": route_delayed_chain,
+        "/delayed/chain5.php": route_delayed_chain,
+        "/delayed/chain6.php": route_delayed_chain,
+        "/delayed/chain7.php": route_delayed_chain,
+        "/delayed/chain8.php": route_delayed_chain,
+        "/delayed/chain9.php": route_delayed_chain,
         "/redir/index.html": route_redir_index,
         "/redir/go.php": route_redir_go,
         "/redir/target.html": route_redir_target,