tests: international crawl tracked the double-encoded behavior

The default.html sub-crawl asserted the pre-fix output: two mojibake
café*.html filenames from double-encoded fetches and 4 errors where
the fix leaves 2 (the caf%e9.html Latin-1 escapes the server really
lacks). Update to the corrected names and counts; the other three
sub-crawls are unchanged. Only runs where ut.httrack.com is reachable
(in-tree builds), which is why just the deb CI job caught it.

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

.github/workflows/codeql.yml

@@ -43,6 +43,12 @@ jobs:
           languages: c-cpp
           build-mode: manual
           queries: security-extended
+          # fopen's umask-controlled 0666 is intended for mirror/cache/log
+          # output; the one credential file (cookies.txt) is kept 0600 on Unix.
+          config: |
+            query-filters:
+              - exclude:
+                  id: cpp/world-writable-file-creation
 
       # Manual build: CodeQL traces the compiler, so build exactly what ships.
       - name: Build

fuzz/Makefile.am

@@ -36,6 +36,7 @@ EXTRA_DIST = README.md run-fuzzers.sh \
 	corpus/unescape/percent.txt \
 	corpus/filters/filter.bin corpus/filters/filter-size.bin \
 	corpus/filters/regress-empty-subject-unique.bin \
+	corpus/filters/redos-star-classes.bin \
 	corpus/url/http-url.txt corpus/url/relative-path.txt \
 	corpus/url/regress-file-empty-path.txt corpus/url/regress-long-path-abort.txt \
 	corpus/header/full-response.txt corpus/header/redirect.txt \

fuzz/corpus/meta/meta-content-first.html

@@ -0,0 +1 @@
+<meta content="text/html; charset=gb2312" http-equiv="content-type">

fuzz/corpus/meta/meta-truncated.html

@@ -0,0 +1 @@
+<meta charset='utf-8

fuzz/corpus/meta/meta-unquoted.html

@@ -0,0 +1 @@
+<meta   charset = utf-8 />

fuzz/fuzz-cachendx.c

@@ -27,7 +27,7 @@ Please visit our Website: http://www.httrack.com
 
 /* Fuzz the cache-index (.ndx) parser: a corrupt or truncated index must not
    walk the length-prefixed cache_brstr/cache_binput scan past the buffer.
-   Mirrors the loader in cache_readex_new (htscache.c). */
+   Mirrors the -#C cache-listing scan (htscoremain.c). */
 #include "fuzz.h"
 #include "htscache.h"
 #include "htslib.h"
@@ -42,9 +42,8 @@ int LLVMFuzzerTestOneInput(const uint8_t *data, size_t size) {
   a += cache_brstr(a, firstline, sizeof(firstline));
   a += cache_brstr(a, firstline, sizeof(firstline));
 
-  /* body: newline-delimited host/file/position triples. The coucal insert
-     the real loader ends with is a separate library's concern; this harness
-     targets the length-prefixed scan that must stay inside the buffer. */
+  /* body: newline-delimited host/file/position triples; the length-prefixed
+     scan must stay inside the buffer */
   while (a != NULL && a < end) {
     char BIGSTK line[HTS_URLMAXSIZE * 2];
     char linepos[256];

fuzz/fuzz-header.c

@@ -50,9 +50,13 @@ int LLVMFuzzerTestOneInput(const uint8_t *data, size_t size) {
   while (p != NULL && *p != '\0') {
     char *nl = strchr(p, '\n');
     size_t n = (nl != NULL) ? (size_t) (nl - p) : strlen(p);
+    size_t i, len = 0;
 
-    memcpy(line, p, n);
-    line[n] = '\0';
+    /* binput drops every '\r' on the wire; mirror it */
+    for (i = 0; i < n; i++)
+      if (p[i] != '\r')
+        line[len++] = p[i];
+    line[len] = '\0';
     if (first) {
       treatfirstline(&r, line);
       first = 0;

fuzz/run-fuzzers.sh

@@ -2,8 +2,8 @@
 # Drive every built harness against its seed corpus.
 #   run-fuzzers.sh <build-fuzz-dir> check       deterministic replay (CI smoke)
 #   run-fuzzers.sh <build-fuzz-dir> [seconds]   timed mutation run (discovery)
-# Replay is crash/leak-only and never mutates, so it can't hit strjoker's
-# catastrophic backtracking; the timed mode can, hence the per-unit -timeout.
+# Replay is crash/leak-only and never mutates; the per-unit -timeout guards
+# both modes against pathological slowdowns (e.g. pre-#501 strjoker).
 set -euo pipefail
 
 srcdir=$(cd "$(dirname "$0")" && pwd)

src/htsback.c

@@ -1646,138 +1646,104 @@ int back_add(struct_back * sback, httrackp * opt, cache_back * cache, const char
       }
     }
     // tester cache
-    if ((strcmp(adr, "file://"))        /* pas fichier */
-        &&((!test) || (cache->type == 1))       /* cache prioritaire, laisser passer en test! */
-        &&((strnotempty(save)) || (strcmp(fil, "/robots.txt") == 0))) { // si en test on ne doit pas utiliser le cache sinon telescopage avec le 302..
-#if HTS_FAST_CACHE
+    if ((strcmp(adr, "file://")) /* pas fichier */
+        && ((!test) ||
+            (cache->type == 1)) /* cache prioritaire, laisser passer en test! */
+        &&
+        ((strnotempty(save)) || (strcmp(fil, "/robots.txt") ==
+                                 0))) { // si en test on ne doit pas utiliser le
+                                        // cache sinon telescopage avec le 302..
       intptr_t hash_pos;
       int hash_pos_return = 0;
-#else
-      char *a = NULL;
-#endif
-#if HTS_FAST_CACHE
+
       if (cache->hashtable) {
-#else
-      if (cache->use) {
-#endif
         char BIGSTK buff[HTS_URLMAXSIZE * 4];
 
-#if HTS_FAST_CACHE
         strcpybuff(buff, adr);
         strcatbuff(buff, fil);
         hash_pos_return = coucal_read(cache->hashtable, buff, &hash_pos);
-#else
-        buff[0] = '\0';
-        strcatbuff(buff, "\n");
-        strcatbuff(buff, adr);
-        strcatbuff(buff, "\n");
-        strcatbuff(buff, fil);
-        strcatbuff(buff, "\n");
-        a = strstr(cache->use, buff);
-#endif
 
-        // Ok, noté en cache->. mais bien présent dans le cache ou sur disque?
-#if HTS_FAST_CACHE
         // negative values when data is not in cache
         if (hash_pos_return < 0) {
-#else
-        if (a) {
-#endif
-          if (!test) {          // non mode test
-#if HTS_FAST_CACHE==0
-            int pos = -1;
+          if (!test) { // not test mode
+            /* note: no check with IS_DELAYED_EXT() enabled - postcheck by
+             * client please! */
+            if (save[0] != '\0' && !IS_DELAYED_EXT(save) &&
+                fsize_utf8(fconv(catbuff, sizeof(catbuff), save)) <=
+                    0) { // final file missing or empty
+              int found = 0;
 
-            a += strlen(buff);
-            sscanf(a, "%d", &pos);      // lire position
-#endif
+              /* It is possible that the file has been moved due to changes in
+               * build structure */
+              {
+                char BIGSTK previous_save[HTS_URLMAXSIZE * 2];
+                htsblk r;
 
-#if HTS_FAST_CACHE==0
-            if (pos < 0) {      // pas de mise en cache data, vérifier existence
-#endif
-              /* note: no check with IS_DELAYED_EXT() enabled - postcheck by client please! */
-              if (save[0] != '\0' && !IS_DELAYED_EXT(save) && fsize_utf8(fconv(catbuff, sizeof(catbuff), save)) <= 0) {  // fichier final n'existe pas ou est vide!
-                int found = 0;
-
-                /* It is possible that the file has been moved due to changes in build structure */
-                {
-                  char BIGSTK previous_save[HTS_URLMAXSIZE * 2];
-                  htsblk r;
-
-                  previous_save[0] = '\0';
-                  r =
-                    cache_readex(opt, cache, adr, fil, /*head */ NULL,
+                previous_save[0] = '\0';
+                r = cache_readex(opt, cache, adr, fil, /*head */ NULL,
                                  /*bound to back[p] (temporary) */
                                  back[p].location_buffer, previous_save, /*ro */
                                  1);
-                  /* Is supposed to be on disk only */
-                  if (r.is_write && previous_save[0] != '\0') {
-                    /* Exists, but with another (old) filename: rename (almost) silently */
-                    if (strcmp(previous_save, save) != 0
-                        && fexist_utf8(fconv(catbuff, sizeof(catbuff), previous_save))) {
-                      rename(fconv(catbuff, sizeof(catbuff), previous_save),
-                             fconv(catbuff2, sizeof(catbuff2), save));
-                      if (fexist_utf8(fconv(catbuff, sizeof(catbuff), save))) {
-                        found = 1;
-                        hts_log_print(opt, LOG_DEBUG,
-                                      "File '%s' has been renamed since last mirror to '%s' ; applying changes",
-                                      previous_save, save);
-                      } else {
-                        hts_log_print(opt, LOG_ERROR,
-                                      "Could not rename '%s' to '%s' ; will have to retransfer it",
-                                      previous_save, save);
-                      }
+                /* Is supposed to be on disk only */
+                if (r.is_write && previous_save[0] != '\0') {
+                  /* Exists, but with another (old) filename: rename (almost)
+                   * silently */
+                  if (strcmp(previous_save, save) != 0 &&
+                      fexist_utf8(
+                          fconv(catbuff, sizeof(catbuff), previous_save))) {
+                    rename(fconv(catbuff, sizeof(catbuff), previous_save),
+                           fconv(catbuff2, sizeof(catbuff2), save));
+                    if (fexist_utf8(fconv(catbuff, sizeof(catbuff), save))) {
+                      found = 1;
+                      hts_log_print(opt, LOG_DEBUG,
+                                    "File '%s' has been renamed since last "
+                                    "mirror to '%s' ; applying changes",
+                                    previous_save, save);
+                    } else {
+                      hts_log_print(opt, LOG_ERROR,
+                                    "Could not rename '%s' to '%s' ; will have "
+                                    "to retransfer it",
+                                    previous_save, save);
                     }
                   }
-                  back[p].location_buffer[0] = '\0';
                 }
+                back[p].location_buffer[0] = '\0';
+              }
 
-                /* Not found ? */
-                if (!found) {
-#if HTS_FAST_CACHE
-                  hash_pos_return = 0;
-#else
-                  a = NULL;
-#endif
-                  // dévalider car non présent sur disque dans structure originale!!!
-                  // sinon, le fichier est ok à priori, mais on renverra un if-modified-since pour
-                  // en être sûr
-                  if (opt->norecatch) { // tester norecatch
-                    if (!fexist_utf8(fconv(catbuff, sizeof(catbuff), save))) {   // fichier existe pas mais déclaré: on l'a effacé
-                      FILE *fp = FOPEN(fconv(catbuff, sizeof(catbuff), save), "wb");
+              /* Not found ? */
+              if (!found) {
+                // invalidate: gone from disk, force a refetch
+                hash_pos_return = 0;
+                if (opt->norecatch) {
+                  if (!fexist_utf8(fconv(
+                          catbuff, sizeof(catbuff),
+                          save))) { // declared but missing: user erased it
+                    FILE *fp =
+                        FOPEN(fconv(catbuff, sizeof(catbuff), save), "wb");
 
-                      if (fp)
-                        fclose(fp);
-                      hts_log_print(opt, LOG_WARNING,
-                                    "Previous file '%s' not found (erased by user ?), ignoring: %s%s",
-                                    save, back[p].url_adr, back[p].url_fil);
-                    }
-                  } else {
+                    if (fp)
+                      fclose(fp);
                     hts_log_print(opt, LOG_WARNING,
-                                  "Previous file '%s' not found (erased by user ?), recatching: %s%s",
+                                  "Previous file '%s' not found (erased by "
+                                  "user ?), ignoring: %s%s",
                                   save, back[p].url_adr, back[p].url_fil);
                   }
+                } else {
+                  hts_log_print(opt, LOG_WARNING,
+                                "Previous file '%s' not found (erased by user "
+                                "?), recatching: %s%s",
+                                save, back[p].url_adr, back[p].url_fil);
                 }
-              }                 // fsize() <= 0
-#if HTS_FAST_CACHE==0
-            }
-#endif
+              }
+            } // fsize() <= 0
           }
         }
         //
       } else {
-#if HTS_FAST_CACHE
         hash_pos_return = 0;
-#else
-        a = NULL;
-#endif
       }
 
-      // Existe pas en cache, ou bien pas de cache présent
-#if HTS_FAST_CACHE
-      if (hash_pos_return) {    // OK existe en cache (et données aussi)!
-#else
-      if (a != NULL) {          // OK existe en cache (et données aussi)!
-#endif
+      if (hash_pos_return) { // in cache, with data
         const int cache_is_prioritary = cache->type == 1
           || opt->state.stop != 0;
         if (cache_is_prioritary) {      // cache prioritaire (pas de test if-modified..)
@@ -1898,7 +1864,8 @@ int back_add(struct_back * sback, httrackp * opt, cache_back * cache, const char
 #endif
         }
       }
-      /* Not in cache ; maybe in temporary cache ? Warning: non-movable "url_sav" */
+      /* Not in cache ; maybe in temporary cache ? Warning: non-movable
+         "url_sav" */
       else if (back_unserialize_ref(opt, adr, fil, &itemback) == 0) {
         const off_t file_size = fsize_utf8(itemback->url_sav);
 

src/htsbauth.c

@@ -307,14 +307,25 @@ int cookie_load(t_cookie * cookie, const char *fpath, const char *name) {
   return -1;
 }
 
-// écrire cookies.txt
-// !=0 : erreur
+/* Write cookies.txt; returns 0 on success. The jar holds live session
+   cookies, so keep it owner-only on Unix (Windows inherits folder ACLs). */
 int cookie_save(t_cookie * cookie, const char *name) {
   char catbuff[CATBUFF_SIZE];
 
   if (strnotempty(cookie->data)) {
     char BIGSTK line[8192];
+#ifdef _WIN32
     FILE *fp = fopen(fconv(catbuff, sizeof(catbuff), name), "wb");
+#else
+    const int fd = open(fconv(catbuff, sizeof(catbuff), name),
+                        O_WRONLY | O_CREAT | O_TRUNC, HTS_PROTECT_FILE);
+    FILE *fp = fd != -1 ? fdopen(fd, "wb") : NULL;
+
+    if (fd != -1 && fp == NULL)
+      close(fd);    /* fdopen failed: don't leak the descriptor */
+    if (fp != NULL) /* O_CREAT's mode skips pre-existing jars: tighten those */
+      (void) fchmod(fd, HTS_PROTECT_FILE);
+#endif
 
     if (fp) {
       char *a = cookie->data;

src/htscache.h

@@ -89,11 +89,6 @@ typedef enum {
    the involved files are absent. */
 void hts_cache_reconcile(httrackp *opt, hts_cache_reconcile_mode mode);
 
-int cache_writedata(FILE * cache_ndx, FILE * cache_dat, const char *str1,
-                    const char *str2, char *outbuff, int len);
-int cache_readdata(cache_back * cache, const char *str1, const char *str2,
-                   char **inbuff, int *len);
-
 void cache_rstr(FILE *fp, char *s, size_t s_size);
 char *cache_rstr_addr(FILE * fp);
 int cache_brstr(char *adr, char *s, size_t s_size);

src/htscache_selftest.c

@@ -78,14 +78,6 @@ static void selftest_open_for_read(cache_back *cache, httrackp *opt) {
 }
 
 static void selftest_close(cache_back *cache) {
-  if (cache->dat != NULL) {
-    fclose(cache->dat);
-    cache->dat = NULL;
-  }
-  if (cache->ndx != NULL) {
-    fclose(cache->ndx);
-    cache->ndx = NULL;
-  }
   if (cache->zipOutput != NULL) {
     zipClose(cache->zipOutput,
              "Created by HTTrack Website Copier (cache self-test)");
@@ -989,26 +981,15 @@ int cache_reconcile_selftest(httrackp *opt, const char *dir) {
   failures +=
       reconcile_expect(opt, "hts-cache/old.zip", SOLID, "promote-zip-noop");
 
-  /* PROMOTE: a pure-legacy old generation is promoted too (was dead when no
-     zip cache existed) */
+  /* PROMOTE: the removed pre-3.31 legacy pair is left alone */
   reconcile_wipe(opt);
   reconcile_put(opt, "hts-cache/old.dat", SOLID);
   reconcile_put(opt, "hts-cache/old.ndx", TINY);
   hts_cache_reconcile(opt, CACHE_RECONCILE_PROMOTE);
-  failures += reconcile_expect(opt, "hts-cache/new.dat", SOLID, "promote-dat");
-  failures += reconcile_expect(opt, "hts-cache/new.ndx", TINY, "promote-dat");
-  failures += reconcile_expect(opt, "hts-cache/old.dat", -1, "promote-dat");
-
-  /* PROMOTE: a half-written legacy new pair is replaced by the old pair */
-  reconcile_wipe(opt);
-  reconcile_put(opt, "hts-cache/new.dat", TINY);
-  reconcile_put(opt, "hts-cache/old.dat", SOLID);
-  reconcile_put(opt, "hts-cache/old.ndx", TINY);
-  hts_cache_reconcile(opt, CACHE_RECONCILE_PROMOTE);
-  failures +=
-      reconcile_expect(opt, "hts-cache/new.dat", SOLID, "promote-dat-partial");
-  failures +=
-      reconcile_expect(opt, "hts-cache/new.ndx", TINY, "promote-dat-partial");
+  failures += reconcile_expect(opt, "hts-cache/new.dat", -1, "promote-dat");
+  failures += reconcile_expect(opt, "hts-cache/new.ndx", -1, "promote-dat");
+  failures += reconcile_expect(opt, "hts-cache/old.dat", SOLID, "promote-dat");
+  failures += reconcile_expect(opt, "hts-cache/old.ndx", TINY, "promote-dat");
 
   /* INTERRUPTED: no lock file, no action */
   reconcile_wipe(opt);
@@ -1058,7 +1039,7 @@ int cache_reconcile_selftest(httrackp *opt, const char *dir) {
   failures +=
       reconcile_expect(opt, "hts-cache/new.zip", MID, "interrupted-bignew");
 
-  /* INTERRUPTED: the legacy pair follows the same size rule (was dead code) */
+  /* INTERRUPTED: the removed pre-3.31 legacy pair is left alone */
   reconcile_wipe(opt);
   reconcile_put(opt, "hts-in_progress.lock", 0);
   reconcile_put(opt, "hts-cache/new.dat", TINY);
@@ -1067,9 +1048,13 @@ int cache_reconcile_selftest(httrackp *opt, const char *dir) {
   reconcile_put(opt, "hts-cache/old.ndx", MID);
   hts_cache_reconcile(opt, CACHE_RECONCILE_INTERRUPTED);
   failures +=
-      reconcile_expect(opt, "hts-cache/new.dat", SOLID, "interrupted-dat");
+      reconcile_expect(opt, "hts-cache/new.dat", TINY, "interrupted-dat");
   failures +=
-      reconcile_expect(opt, "hts-cache/new.ndx", MID, "interrupted-dat");
+      reconcile_expect(opt, "hts-cache/new.ndx", TINY, "interrupted-dat");
+  failures +=
+      reconcile_expect(opt, "hts-cache/old.dat", SOLID, "interrupted-dat");
+  failures +=
+      reconcile_expect(opt, "hts-cache/old.ndx", MID, "interrupted-dat");
 
   /* ROLLBACK: the old zip generation is restored (a zip cache used to lose
      its only good generation here) */
@@ -1089,17 +1074,18 @@ int cache_reconcile_selftest(httrackp *opt, const char *dir) {
   failures += reconcile_expect(opt, "hts-cache/new.lst", MID, "rollback-lst");
   failures += reconcile_expect(opt, "hts-cache/new.txt", MID, "rollback-txt");
 
-  /* ROLLBACK: full legacy generation incl. sidecars (historical behavior) */
+  /* ROLLBACK: sidecars restored, the removed pre-3.31 pair left alone */
   reconcile_wipe(opt);
   reconcile_put(opt, "hts-cache/new.dat", TINY);
-  reconcile_put(opt, "hts-cache/new.ndx", TINY);
   reconcile_put(opt, "hts-cache/old.dat", SOLID);
   reconcile_put(opt, "hts-cache/old.ndx", MID);
   reconcile_put(opt, "hts-cache/old.lst", MID);
   reconcile_put(opt, "hts-cache/old.txt", MID);
   hts_cache_reconcile(opt, CACHE_RECONCILE_ROLLBACK);
-  failures += reconcile_expect(opt, "hts-cache/new.dat", SOLID, "rollback-dat");
-  failures += reconcile_expect(opt, "hts-cache/new.ndx", MID, "rollback-dat");
+  failures += reconcile_expect(opt, "hts-cache/new.dat", TINY, "rollback-dat");
+  failures += reconcile_expect(opt, "hts-cache/new.ndx", -1, "rollback-dat");
+  failures += reconcile_expect(opt, "hts-cache/old.dat", SOLID, "rollback-dat");
+  failures += reconcile_expect(opt, "hts-cache/old.ndx", MID, "rollback-dat");
   failures += reconcile_expect(opt, "hts-cache/new.lst", MID, "rollback-dat");
   failures += reconcile_expect(opt, "hts-cache/new.txt", MID, "rollback-dat");
 
@@ -1113,6 +1099,101 @@ int cache_reconcile_selftest(httrackp *opt, const char *dir) {
   return failures;
 }
 
+/* The pre-3.31 .dat/.ndx import was removed: cache_init must refuse the
+   legacy pair, leave the files in place, and not flag an update run. */
+int cache_legacy_refused_selftest(httrackp *opt, const char *dir) {
+  int failures = 0;
+  int variant;
+
+  selftest_tag = "cache-legacy";
+  golden_setup(opt, dir);
+#ifdef _WIN32
+  mkdir(reconcile_st_path(opt, "hts-cache"));
+#else
+  mkdir(reconcile_st_path(opt, "hts-cache"), HTS_PROTECT_FOLDER);
+#endif
+
+  /* variant 0: old.dat/old.ndx (--update layout); 1: new.dat/new.ndx (ro) */
+  for (variant = 0; variant < 2; variant++) {
+    cache_back cache;
+    const char *const dat =
+        variant == 0 ? "hts-cache/old.dat" : "hts-cache/new.dat";
+    const char *const ndx =
+        variant == 0 ? "hts-cache/old.ndx" : "hts-cache/new.ndx";
+
+    reconcile_wipe(opt);
+    reconcile_put(opt, dat, 4096);
+    reconcile_put(opt, ndx, 4096);
+    opt->is_update = 0;
+    selftest_open(&cache, opt, /*ro*/ variant == 1);
+    if (cache_readable(&cache)) {
+      printf("cache-legacy: variant %d imported a removed format\n", variant);
+      failures++;
+    }
+    if (coucal_nitems(cache.hashtable) != 0) { /* no phantom index entries */
+      printf("cache-legacy: variant %d imported index entries\n", variant);
+      failures++;
+    }
+    if (opt->is_update) {
+      printf("cache-legacy: variant %d flagged an update\n", variant);
+      failures++;
+    }
+    if (fsize(reconcile_st_path(opt, dat)) != 4096 ||
+        fsize(reconcile_st_path(opt, ndx)) != 4096) {
+      printf("cache-legacy: variant %d touched the legacy files\n", variant);
+      failures++;
+    }
+    if (cache.lst != NULL) {
+      fclose(cache.lst);
+      cache.lst = NULL;
+    }
+    if (cache.txt != NULL) {
+      fclose(cache.txt);
+      cache.txt = NULL;
+    }
+    selftest_close(&cache);
+  }
+
+  /* a healthy zip must win over stray legacy files, with no refusal */
+  {
+    cache_back cache;
+    const char *const body = "<html>zip wins</html>";
+
+    reconcile_wipe(opt);
+    selftest_open_for_write(&cache, opt);
+    store_entry(opt, &cache, "example.com", "/", "example.com/index.html", 200,
+                "OK", "text/html", "utf-8", "", "", "", "", body, strlen(body));
+    selftest_close(&cache);
+    reconcile_put(opt, "hts-cache/old.ndx", 4096);
+    reconcile_put(opt, "hts-cache/old.dat", 4096);
+    selftest_open_for_read(&cache, opt);
+    if (!cache_readable(&cache)) {
+      printf("cache-legacy: stray legacy files shadowed the zip cache\n");
+      failures++;
+    }
+    failures +=
+        check_entry(opt, &cache, "example.com", "/", 200, "OK", "text/html",
+                    "utf-8", "", "", "", "", body, strlen(body));
+    if (fsize(reconcile_st_path(opt, "hts-cache/old.ndx")) != 4096 ||
+        fsize(reconcile_st_path(opt, "hts-cache/old.dat")) != 4096) {
+      printf("cache-legacy: zip-wins pass touched the legacy files\n");
+      failures++;
+    }
+    if (cache.lst != NULL) {
+      fclose(cache.lst);
+      cache.lst = NULL;
+    }
+    if (cache.txt != NULL) {
+      fclose(cache.txt);
+      cache.txt = NULL;
+    }
+    selftest_close(&cache);
+  }
+
+  reconcile_wipe(opt);
+  return failures;
+}
+
 /* --- read-side corruption injection --------------------------------------- */
 
 /* canary read back intact after each corruption; victim gets the byte surgery

src/htscache_selftest.h

@@ -60,6 +60,10 @@ int cache_write_failure_selftest(httrackp *opt, const char *dir);
    under <dir>. Returns the failed-check count. */
 int cache_reconcile_selftest(httrackp *opt, const char *dir);
 
+/* Verify cache_init refuses a pre-3.31 .dat/.ndx cache without touching it.
+   Returns the number of failed checks (0 = pass). */
+int cache_legacy_refused_selftest(httrackp *opt, const char *dir);
+
 /* Inject read-side corruption (zip byte surgery: bad size, header, deflate)
    under <dir> and assert every case degrades to STATUSCODE_INVALID without
    tainting a sibling entry. */

src/htscharset.c

@@ -84,7 +84,7 @@ static int hts_equalsAlphanum(const char *a, const char *b) {
 
 /* Copy the memory region [s .. s + size - 1 ] as a \0-terminated string. */
 static char *hts_stringMemCopy(const char *s, size_t size) {
-  char *dest = malloc(size + 1);
+  char *dest = malloct(size + 1);
 
   if (dest != NULL) {
     memcpy(dest, s, size);
@@ -549,42 +549,6 @@ char *hts_convertStringFromUTF8(const char *s, size_t size, const char *charset)
 
 #endif
 
-HTS_STATIC char *hts_getCharsetFromContentType(const char *mime) {
-  /* text/html; charset=utf-8 */
-  const char *const charset = "charset";
-  char *pos = strstr(mime, charset);
-
-  if (pos != NULL) {
-    /* Skip spaces */
-    int eq = 0;
-
-    for(pos += strlen(charset);
-        *pos == ' ' || *pos == '=' || *pos == '"' || *pos == '\''; pos++) {
-      if (*pos == '=') {
-        eq = 1;
-      }
-    }
-    if (eq == 1) {
-      int len;
-
-      for(len = 0;
-          pos[len] == ' ' || pos[len] == ';' || pos[len] == '"' || *pos == '\'';
-          pos++) ;
-      if (len != 0) {
-        char *const s = malloc(len + 1);
-        int i;
-
-        for(i = 0; i < len; i++) {
-          s[i] = pos[i];
-        }
-        s[len] = '\0';
-        return s;
-      }
-    }
-  }
-  return NULL;
-}
-
 #ifdef _WIN32
 #define strcasecmp(a,b) stricmp(a,b)
 #define strncasecmp(a,b,n) strnicmp(a,b,n)
@@ -644,58 +608,106 @@ int hts_isCharsetUTF8(const char *charset) {
          || strcasecmp(charset, "utf8") == 0 );
 }
 
+/* Extract X from a "text/html; charset=X" content= attribute value. */
+static char *charset_from_content(const char *s, size_t len) {
+  size_t i;
+
+  for (i = 0; i + 7 < len; i++) {
+    if ((i == 0 || is_space(s[i - 1]) || s[i - 1] == ';') &&
+        strncasecmp(&s[i], "charset", 7) == 0 && is_space_or_equal(s[i + 7])) {
+      size_t j, val;
+
+      for (j = i + 7; j < len && is_space_or_equal_or_quote(s[j]); j++)
+        ;
+      for (val = j; j < len && s[j] != '"' && s[j] != '\'' && s[j] != ';' &&
+                    !is_space(s[j]);
+           j++)
+        ;
+      if (j != val) {
+        return hts_stringMemCopy(&s[val], j - val);
+      }
+    }
+  }
+  return NULL;
+}
+
 char *hts_getCharsetFromMeta(const char *html, size_t size) {
-  int i;
+  size_t i;
 
-  /* <META HTTP-EQUIV="CONTENT-TYPE" CONTENT="text/html; charset=utf-8" > */
-  for(i = 0; i < size; i++) {
-    if (html[i] == '<' && strncasecmp(&html[i + 1], "meta", 4) == 0
-        && is_space(html[i + 5])) {
-      /* Skip spaces */
-      for(i += 5; is_space(html[i]); i++) ;
-      if (strncasecmp(&html[i], "HTTP-EQUIV", 10) == 0
-          && is_space_or_equal(html[i + 10])) {
-        for(i += 10; is_space_or_equal_or_quote(html[i]); i++) ;
-        if (strncasecmp(&html[i], "CONTENT-TYPE", 12) == 0) {
-          for(i += 12; is_space_or_equal_or_quote(html[i]); i++) ;
-          if (strncasecmp(&html[i], "CONTENT", 7) == 0
-              && is_space_or_equal(html[i + 7])) {
-            for(i += 7; is_space_or_equal_or_quote(html[i]); i++) ;
-            /* Skip content-type */
-            for(;
-                i < size && html[i] != ';' && html[i] != '"' && html[i] != '\'';
-                i++) ;
-            /* Expect charset attribute here */
-            if (html[i] == ';') {
-              for(i++; is_space(html[i]); i++) ;
-              /* Look for charset */
-              if (strncasecmp(&html[i], "charset", 7) == 0
-                  && is_space_or_equal(html[i + 7])) {
-                int len;
+  /* HTML5 <meta charset=X> or legacy
+     <meta http-equiv="Content-Type" content="text/html; charset=X">,
+     attributes in any order; first resolvable tag wins. */
+  for (i = 0; i + 6 < size; i++) {
+    size_t j;
+    const char *equiv = NULL, *content = NULL;
+    size_t equiv_len = 0, content_len = 0;
 
-                for(i += 7; is_space_or_equal(html[i]) || html[i] == '\'';
-                    i++) ;
-                /* Charset */
-                for(len = 0;
-                    i + len < size && html[i + len] != '"'
-                    && html[i + len] != '\'' && html[i + len] != ' '; len++) ;
-                /* No error ? */
-                if (len != 0 && i < size) {
-                  char *const s = malloc(len + 1);
-                  int j;
+    if (html[i] != '<' || strncasecmp(&html[i + 1], "meta", 4) != 0 ||
+        !is_space(html[i + 5])) {
+      continue;
+    }
+    /* Attribute scan, strictly bounded by size. */
+    for (j = i + 6; j < size && html[j] != '>';) {
+      size_t name, name_len, val = 0, val_len = 0;
 
-                  for(j = 0; j < len; j++) {
-                    s[j] = html[i + j];
-                  }
-                  s[len] = '\0';
-                  return s;
-                }
-              }
-            }
+      if (is_space(html[j]) || html[j] == '/') {
+        j++;
+        continue;
+      }
+      for (name = j; j < size && html[j] != '=' && html[j] != '>' &&
+                     html[j] != '/' && !is_space(html[j]);
+           j++)
+        ;
+      name_len = j - name;
+      for (; j < size && is_space(html[j]); j++)
+        ;
+      if (j < size && html[j] == '=') {
+        for (j++; j < size && is_space(html[j]); j++)
+          ;
+        if (j < size && (html[j] == '"' || html[j] == '\'')) {
+          const char quote = html[j++];
+
+          for (val = j; j < size && html[j] != quote; j++)
+            ;
+          if (j >= size) /* unterminated quote: drop the tag */
+            break;
+          val_len = j++ - val;
+        } else {
+          /* unquoted: ends at whitespace or '>' only (HTML5 prescan); '/'
+             belongs to the value except as the tail of a self-close */
+          for (val = j; j < size && !is_space(html[j]) && html[j] != '>'; j++)
+            ;
+          val_len = j - val;
+          if (val_len != 0 && j < size && html[j] == '>' &&
+              html[val + val_len - 1] == '/') {
+            val_len--;
           }
         }
       }
+      /* first occurrence of each attribute wins, as in browsers */
+      if (val_len != 0) {
+        if (name_len == 7 && strncasecmp(&html[name], "charset", 7) == 0) {
+          return hts_stringMemCopy(&html[val], val_len);
+        } else if (equiv == NULL && name_len == 10 &&
+                   strncasecmp(&html[name], "http-equiv", 10) == 0) {
+          equiv = &html[val];
+          equiv_len = val_len;
+        } else if (content == NULL && name_len == 7 &&
+                   strncasecmp(&html[name], "content", 7) == 0) {
+          content = &html[val];
+          content_len = val_len;
+        }
+      }
     }
+    if (equiv_len == 12 && strncasecmp(equiv, "content-type", 12) == 0 &&
+        content != NULL) {
+      char *const s = charset_from_content(content, content_len);
+
+      if (s != NULL) {
+        return s;
+      }
+    }
+    i = j;
   }
   return NULL;
 }
@@ -1140,17 +1152,45 @@ int hts_isStringUTF8(const char *s, size_t size) {
   const unsigned char *const data = (const unsigned char*) s;
   size_t i;
 
+  /* Strict RFC 3629: reject overlongs, surrogates, > U+10FFFF, 5/6-byte. */
   for(i = 0 ; i < size ; ) {
-    /* Reader: can read bytes up to j */
-#define RD ( i < size ? data[i++] : -1 )
+    const unsigned char c = data[i];
+    size_t len, k;
+    hts_UCS4 uc, min;
 
-    /* Writer: a malformed sequence means the string is not UTF-8 (return 0) */
-#define WR(C) if ((C) == -1) { return 0; }
+    if (c < 0x80) {
+      i++;
+      continue;
+    } else if (c >= 0xc2 && c <= 0xdf) {
+      len = 2;
+      min = 0x80;
+      uc = c & 0x1f;
+    } else if (c >= 0xe0 && c <= 0xef) {
+      len = 3;
+      min = 0x800;
+      uc = c & 0x0f;
+    } else if (c >= 0xf0 && c <= 0xf4) {
+      len = 4;
+      min = 0x10000;
+      uc = c & 0x07;
+    } else { /* continuation byte, overlong C0/C1, or F5..FF lead */
+      return 0;
+    }
+    if (size - i < len) {
+      return 0;
+    }
+    for (k = 1; k < len; k++) {
+      const unsigned char cc = data[i + k];
 
-    /* Read Unicode character. */
-    READ_UNICODE(RD, WR);
-#undef RD
-#undef WR
+      if ((cc & 0xc0) != 0x80) {
+        return 0;
+      }
+      uc = (uc << 6) | (cc & 0x3f);
+    }
+    if (uc < min || uc > 0x10FFFF || (uc >= 0xD800 && uc <= 0xDFFF)) {
+      return 0;
+    }
+    i += len;
   }
 
   return 1;

src/htscharset.h

@@ -76,7 +76,8 @@ extern char *hts_convertStringIDNAToUTF8(const char *s, size_t size);
 extern int hts_isStringIDNA(const char *s, size_t size);
 
 /**
- * Extract the charset from the HTML buffer "html"
+ * Extract the <meta> charset from the HTML buffer "html" (HTML5 charset= or
+ * legacy http-equiv form). Returns a malloc'ed string, or NULL if none.
  **/
 extern char *hts_getCharsetFromMeta(const char *html, size_t size);
 
@@ -86,7 +87,8 @@ extern char *hts_getCharsetFromMeta(const char *html, size_t size);
 extern int hts_isStringAscii(const char *s, size_t size);
 
 /**
- * Is the given string an UTF-8 string ?
+ * Is the given string valid UTF-8 ? Strict RFC 3629: overlong forms,
+ * surrogates, codepoints above U+10FFFF and 5/6-byte sequences are rejected.
  **/
 extern int hts_isStringUTF8(const char *s, size_t size);
 

src/htsconfig.h

@@ -76,9 +76,6 @@ Please visit our Website: http://www.httrack.com
 // always direct-to-disk (0/1)
 #define HTS_DIRECTDISK_ALWAYS 1
 
-// fast cache (build hash table)
-#define HTS_FAST_CACHE 1
-
 // le > peut être considéré comme un tag de fermeture de commentaire (<!-- > est
 // valide)
 #define GT_ENDS_COMMENT 1

src/htscore.c

@@ -134,18 +134,6 @@ RUN_CALLBACK0(opt, end); \
     back_delete_all(opt, &cache, sback);                                       \
     back_free(&sback);                                                         \
     checkrobots_free(&robots);                                                 \
-    if (cache.use) {                                                           \
-      freet(cache.use);                                                        \
-      cache.use = NULL;                                                        \
-    }                                                                          \
-    if (cache.dat) {                                                           \
-      fclose(cache.dat);                                                       \
-      cache.dat = NULL;                                                        \
-    }                                                                          \
-    if (cache.ndx) {                                                           \
-      fclose(cache.ndx);                                                       \
-      cache.ndx = NULL;                                                        \
-    }                                                                          \
     if (cache.zipOutput) {                                                     \
       zipClose(cache.zipOutput,                                                \
                "Created by HTTrack Website Copier/" HTTRACK_VERSION);          \
@@ -155,10 +143,6 @@ RUN_CALLBACK0(opt, end); \
       unzClose(cache.zipInput);                                                \
       cache.zipInput = NULL;                                                   \
     }                                                                          \
-    if (cache.olddat) {                                                        \
-      fclose(cache.olddat);                                                    \
-      cache.olddat = NULL;                                                     \
-    }                                                                          \
     if (cache.lst) {                                                           \
       fclose(cache.lst);                                                       \
       cache.lst = opt->state.strc.lst = NULL;                                  \
@@ -488,32 +472,11 @@ void hts_finish_makeindex(httrackp *opt, int *makeindex_done,
   *makeindex_done = 1;
 }
 
-/* Flush the parsed HTML output buffer to disk, skipping the rewrite when the
- * on-disk MD5 is unchanged. */
+/* Flush the parsed HTML output buffer to disk. */
 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) {
-  char digest[32 + 2];
-  off_t fsize_old =
-      fsize(fconv(OPT_GET_BUFF(opt), OPT_GET_BUFF_SIZE(opt), save));
-  int ok = 0;
-
-  digest[0] = '\0';
-  domd5mem(ht_buff, ht_len, digest, 1);
-  if (fsize_old == (off_t) ht_len) {
-    int mlen = 0;
-    char *mbuff;
-
-    cache_readdata(cache, "//[HTML-MD5]//", save, &mbuff, &mlen);
-    if (mlen)
-      mbuff[mlen] = '\0';
-    if ((mlen == 32) && (strcmp(((mbuff != NULL) ? mbuff : ""), digest) == 0)) {
-      ok = 1;
-      hts_log_print(opt, LOG_DEBUG, "File not re-written (md5): %s", save);
-    }
-    freet(mbuff);
-  }
-  if (!ok) {
+  {
     file_notify(opt, adr, fil, save, 1, 1, r->notmodified);
     *fp = filecreate(&opt->state.strc, save);
     if (*fp) {
@@ -545,13 +508,7 @@ void hts_finish_html_file(httrackp *opt, cache_back *cache, htsblk *r,
       if (fcheck)
         hts_log_print(opt, LOG_ERROR, "* * Fatal write error, giving up");
     }
-  } else {
-    file_notify(opt, adr, fil, save, 0, 0, r->notmodified);
-    filenote(&opt->state.strc, save, NULL);
   }
-  if (cache->ndx)
-    cache_writedata(cache->ndx, cache->dat, "//[HTML-MD5]//", save, digest,
-                    (int) strlen(digest));
 }
 
 /* does it look like XML ? (SVG et al.) */

src/htscore.h

@@ -193,8 +193,7 @@ struct cache_back {
   /* */
   int type;
   int ro;                   /**< read-only: no new cache is written */
-  FILE *dat, *ndx, *olddat; /**< new data, new index, old data files */
-  char *use;                /**< in-memory list of cached adr+fil keys */
+
   FILE *lst;                /**< file list, used for purge */
   FILE *txt;                /**< human-readable file list (info) */
   char lastmodified[256];
@@ -288,12 +287,12 @@ struct filecreate_params {
 
 /* True if a new cache is being written (plain or zip backend). */
 HTS_STATIC int cache_writable(cache_back * cache) {
-  return (cache != NULL && (cache->dat != NULL || cache->zipOutput != NULL));
+  return (cache != NULL && cache->zipOutput != NULL);
 }
 
 /* True if an old cache is available to read (plain or zip backend). */
 HTS_STATIC int cache_readable(cache_back * cache) {
-  return (cache != NULL && (cache->olddat != NULL || cache->zipInput != NULL));
+  return (cache != NULL && cache->zipInput != NULL);
 }
 
 #endif

src/htsfilters.c

@@ -48,6 +48,7 @@ Please visit our Website: http://www.httrack.com
 #include "htsbase.h"
 #include "htslib.h"
 #include <ctype.h>
+#include <stdint.h>
 /* END specific definitions */
 
 // à partir d'un tableau de {"+*.toto","-*.zip","+*.tata"} définit si nom est autorisé
@@ -120,16 +121,81 @@ int fa_strjoker_dual(int type, char **filters, int nfil, const char *nom1,
   return use1 ? jok1 : jok2;
 }
 
-// supercomparateur joker (tm)
-// compare a et b (b=avec joker dedans), case insensitive [voir CI]
-// renvoi l'adresse de la première lettre de la chaine
-// (càd *[..]toto.. renvoi adresse de toto dans la chaine)
-// accepte les délires du genre www.*.*/ * / * truc*.*
-// cet algo est 'un peu' récursif mais ne consomme pas trop de tm
-// * = toute lettre
-// --?-- : spécifique à HTTrack et aux ?
+/* Failure memo for the recursive matcher: one bit per (chaine, joker) offset
+   pair keeps star-heavy patterns polynomial instead of exponential (#501). */
+typedef struct strjoker_memo {
+  const char *chaine0, *joker0; /* offsets are relative to these bases */
+  size_t stride;                /* strlen(joker0) + 1 */
+  unsigned char *failed;        /* failed-pair bitmap; NULL: no memo */
+} strjoker_memo;
+
+static const char *strjoker_impl(const strjoker_memo *memo, const char *chaine,
+                                 const char *joker, LLint *size,
+                                 int *size_flag);
+
+/* A pair that failed once fails forever (*size is only written on the success
+   path), so record NULL results and cut the re-exploration. */
+static const char *strjoker_rec(const strjoker_memo *memo, const char *chaine,
+                                const char *joker, LLint *size,
+                                int *size_flag) {
+  size_t bit = 0;
+  const char *adr;
+
+  if (memo->failed) {
+    bit = (size_t) (chaine - memo->chaine0) * memo->stride +
+          (size_t) (joker - memo->joker0);
+    if (memo->failed[bit >> 3] & (unsigned char) (1u << (bit & 7)))
+      return NULL;
+  }
+  adr = strjoker_impl(memo, chaine, joker, size, size_flag);
+  if (adr == NULL && memo->failed)
+    memo->failed[bit >> 3] |= (unsigned char) (1u << (bit & 7));
+  return adr;
+}
+
+// wildcard comparator: match chaine against joker (pattern), case-insensitive
+// returns the address of the first matched letter past any leading joker
+// (ie. *[..]toto.. returns the address of toto within chaine), NULL on mismatch
 HTS_INLINE const char *strjoker(const char *chaine, const char *joker,
                                 LLint *size, int *size_flag) {
+  strjoker_memo memo = {chaine, joker, 0, NULL};
+  unsigned char stackbits[2048];
+  hts_boolean onheap = HTS_FALSE;
+  const char *adr;
+
+  if (chaine != NULL && joker != NULL) {
+    const size_t n1 = strlen(chaine) + 1;
+
+    memo.stride = strlen(joker) + 1;
+    if (n1 <= (SIZE_MAX - 7) / memo.stride) { // overflow-safe bitmap size
+      const size_t bytes = (n1 * memo.stride + 7) / 8;
+
+      if (bytes <= sizeof(stackbits)) {
+        memset(stackbits, 0, bytes);
+        memo.failed = stackbits;
+      } else {
+        memo.failed = (unsigned char *) calloct(bytes, 1);
+        onheap = memo.failed != NULL ? HTS_TRUE : HTS_FALSE;
+      }
+    }
+  }
+  adr = strjoker_rec(&memo, chaine, joker, size, size_flag);
+  if (onheap)
+    freet(memo.failed);
+  return adr;
+}
+
+/* Test-only oracle: the same matcher without the failure memo. */
+const char *strjoker_nomemo(const char *chaine, const char *joker, LLint *size,
+                            int *size_flag) {
+  const strjoker_memo memo = {chaine, joker, 0, NULL};
+
+  return strjoker_rec(&memo, chaine, joker, size, size_flag);
+}
+
+static const char *strjoker_impl(const strjoker_memo *memo, const char *chaine,
+                                 const char *joker, LLint *size,
+                                 int *size_flag) {
   if (strnotempty(joker) == 0) {        // fin de chaine joker
     if (strnotempty(chaine) == 0)       // fin aussi pour la chaine: ok
       return chaine;
@@ -303,7 +369,7 @@ HTS_INLINE const char *strjoker(const char *chaine, const char *joker,
 
       // tester sans le joker (pas ()+ mais ()*)
       if (!unique) {
-        if ((adr = strjoker(chaine, joker + jmp, size, size_flag))) {
+        if ((adr = strjoker_rec(memo, chaine, joker + jmp, size, size_flag))) {
           return adr;
         }
       }
@@ -315,7 +381,8 @@ HTS_INLINE const char *strjoker(const char *chaine, const char *joker,
         max = strnotempty(chaine) ? 1 : 0; /* empty chaine: no char to eat */
       while(i < (int) max) {
         if (pass[(int) (unsigned char) chaine[i]]) {    // caractère autorisé
-          if ((adr = strjoker(chaine + i + 1, joker + jmp, size, size_flag))) {
+          if ((adr = strjoker_rec(memo, chaine + i + 1, joker + jmp, size,
+                                  size_flag))) {
             return adr;
           }
           i++;
@@ -325,7 +392,8 @@ HTS_INLINE const char *strjoker(const char *chaine, const char *joker,
 
       // tester chaîne vide
       if (i != max + 2)         // avant c'est ok
-        if ((adr = strjoker(chaine + max, joker + jmp, size, size_flag)))
+        if ((adr = strjoker_rec(memo, chaine + max, joker + jmp, size,
+                                size_flag)))
           return adr;
 
       return NULL;              // perdu
@@ -347,7 +415,7 @@ HTS_INLINE const char *strjoker(const char *chaine, const char *joker,
       // comparaison ok?
       if (ok) {
         // continuer la comparaison.
-        if (strjoker(chaine + jmp, joker + jmp, size, size_flag))
+        if (strjoker_rec(memo, chaine + jmp, joker + jmp, size, size_flag))
           return chaine;        // retourner 1e lettre
       }
 

src/htsfilters.h

@@ -48,6 +48,10 @@ int fa_strjoker_dual(int type, char **filters, int nfil, const char *nom1,
                      const char *nom2, LLint *size, int *size_flag, int *depth);
 HTS_INLINE const char *strjoker(const char *chaine, const char *joker, LLint * size,
                           int *size_flag);
+/* strjoker() without the failure memo (exponential worst case); test-only
+   oracle for the memoized matcher. */
+const char *strjoker_nomemo(const char *chaine, const char *joker, LLint *size,
+                            int *size_flag);
 const char *strjokerfind(const char *chaine, const char *joker);
 #endif
 

src/htsglobal.h

@@ -422,9 +422,10 @@ typedef int T_SOC;
 #endif
 
 /* Permission bits for created folders and files (mkdir and chmod).
-   PROTECT_FOLDER is owner-only. With HTS_ACCESS set (the default) the ACCESS_
-   modes also grant group/other read; otherwise they stay owner-only. */
+   PROTECT_FOLDER/FILE are owner-only. With HTS_ACCESS set (the default) the
+   ACCESS_ modes also grant group/other read; otherwise they stay owner-only. */
 #define HTS_PROTECT_FOLDER (S_IRUSR | S_IWUSR | S_IXUSR)
+#define HTS_PROTECT_FILE (S_IRUSR | S_IWUSR)
 
 #if HTS_ACCESS
 #define HTS_ACCESS_FILE (S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH)

src/htsparse.c

@@ -2007,7 +2007,7 @@ int htsparse(htsmoduleStruct * str, htsmoduleStructExtended * stre) {
 
                 /* Unescape/escape %20 and other   */
                 {
-                  // Note: always true (iso-8859-1 as default)
+                  // NULL when UTF-8 conversion is off (-%T0)
                   const char *const charset = str->page_charset_;
                   const int hasCharset = charset != NULL 
                     && *charset != '\0';
@@ -2030,11 +2030,12 @@ int htsparse(htsmoduleStruct * str, htsmoduleStructExtended * stre) {
 
                   // Force to encode non-printable chars (should never happend)
                   escape_remove_control(lien);
-                  
-                  // charset conversion for the URI filename, 
-                  // and not already UTF-8
-                  // (note: not for the query string!)
-                  if (hasCharset && !hts_isCharsetUTF8(charset)) {
+
+                  // charset conversion for the URI filename (not the query
+                  // string), unless the bytes already are valid UTF-8:
+                  // converting those would double-encode them (#180)
+                  if (hasCharset && !hts_isCharsetUTF8(charset) &&
+                      !hts_isStringUTF8(lien, strlen(lien))) {
                     char *const s = hts_convertStringToUTF8(lien, strlen(lien), charset);
                     if (s != NULL) {
                       hts_log_print(opt, LOG_DEBUG,

src/htsselftest.c

@@ -621,6 +621,105 @@ static int st_filtersize(httrackp *opt, int argc, char **argv) {
   return 0;
 }
 
+/* Mime-type filter verdict via fa_strjoker(type=1): only mime: rules apply. */
+static int st_filtermime(httrackp *opt, int argc, char **argv) {
+  int verdict;
+
+  (void) opt;
+  if (argc < 2) {
+    fprintf(stderr, "filtermime: needs <mime> <filter> [filter...]\n");
+    return 1;
+  }
+  verdict = fa_strjoker(1, &argv[1], argc - 1, argv[0], NULL, NULL, NULL);
+  printf("verdict=%s\n", verdict > 0   ? "allowed"
+                         : verdict < 0 ? "forbidden"
+                                       : "unknown");
+  return 0;
+}
+
+/* SplitMix64: deterministic, platform-independent case generator. */
+static uint64_t st_mix64(uint64_t *state) {
+  uint64_t z = (*state += UINT64_C(0x9E3779B97F4A7C15));
+
+  z = (z ^ (z >> 30)) * UINT64_C(0xBF58476D1CE4B5B9);
+  z = (z ^ (z >> 27)) * UINT64_C(0x94D049BB133111EB);
+  return z ^ (z >> 31);
+}
+
+/* Differential test: memoized strjoker vs the no-memo oracle on seeded random
+   pattern/subject/size cases must agree on result, *size and *size_flag. */
+static int st_filtermemo(httrackp *opt, int argc, char **argv) {
+  static const char *const pieces[] = {
+      "a",       "b",        "A",    "c",      ".",       "/",
+      "?",       "*",        "*[a]", "*[ab]",  "*[a-c]",  "*[A-Z]",
+      "*[<5]",   "*[>5]",    "*(a)", "*(a,b)", "*[file]", "*[path]",
+      "*[name]", "*[param]", "*[]",  "*[\\a]", "*[a-"};
+  static const char subject_chars[] = "abAc./?";
+  uint64_t rng = UINT64_C(0x501);
+  int iters = 20000, matched = 0, unmatched = 0;
+  int it;
+
+  (void) opt;
+  if (argc >= 1)
+    sscanf(argv[0], "%d", &iters);
+  for (it = 0; it < iters; it++) {
+    char pat[64], str[16];
+    size_t pl = 0;
+    const int npieces = (int) (st_mix64(&rng) % 5);
+    const int slen = (int) (st_mix64(&rng) % 13);
+    const int szsel = (int) (st_mix64(&rng) % 4); /* none/unknown/3/20 */
+    LLint sz1, sz2, off1, off2;
+    int i, flag1 = 0, flag2 = 0;
+    char *hpat, *hstr;
+    const char *adr;
+
+    for (i = 0; i < npieces; i++) {
+      const char *p =
+          pieces[st_mix64(&rng) % (sizeof(pieces) / sizeof(pieces[0]))];
+      const size_t len = strlen(p);
+
+      if (len >= sizeof(pat) - pl)
+        break;
+      memcpy(pat + pl, p, len);
+      pl += len;
+    }
+    pat[pl] = '\0';
+    for (i = 0; i < slen; i++)
+      str[i] = subject_chars[st_mix64(&rng) % (sizeof(subject_chars) - 1)];
+    str[slen] = '\0';
+    sz1 = sz2 = (szsel <= 1) ? -1 : (szsel == 2) ? 3 : 20;
+    /* exact-size heap copies per run so a sanitizer traps any over-read */
+    hpat = strdupt(pat);
+    hstr = strdupt(str);
+    adr = strjoker(hstr, hpat, szsel ? &sz1 : NULL, szsel ? &flag1 : NULL);
+    off1 = adr != NULL ? (LLint) (adr - hstr) : -1;
+    freet(hpat);
+    freet(hstr);
+    hpat = strdupt(pat);
+    hstr = strdupt(str);
+    adr =
+        strjoker_nomemo(hstr, hpat, szsel ? &sz2 : NULL, szsel ? &flag2 : NULL);
+    off2 = adr != NULL ? (LLint) (adr - hstr) : -1;
+    freet(hpat);
+    freet(hstr);
+    if (off1 != off2 || sz1 != sz2 || flag1 != flag2) {
+      printf("filtermemo MISMATCH pat=[%s] str=[%s] szsel=%d: "
+             "off %d/%d size %d/%d flag %d/%d\n",
+             pat, str, szsel, (int) off1, (int) off2, (int) sz1, (int) sz2,
+             flag1, flag2);
+      return 1;
+    }
+    if (off1 >= 0)
+      matched++;
+    else
+      unmatched++;
+  }
+  /* both polarities must actually occur or the test proves nothing */
+  assertf(matched > 0 && unmatched > 0);
+  printf("filtermemo: %d cases OK\n", iters);
+  return 0;
+}
+
 /* Merged two-form filter verdict via fa_strjoker_dual (see htsfilters.h). */
 static int st_filterdual(httrackp *opt, int argc, char **argv) {
   int depth = -1, verdict;
@@ -689,6 +788,30 @@ static int st_charset(httrackp *opt, int argc, char **argv) {
   return 0;
 }
 
+static int st_metacharset(httrackp *opt, int argc, char **argv) {
+  char *s;
+
+  (void) opt;
+  if (argc < 1) {
+    fprintf(stderr, "metacharset: needs an html string\n");
+    return 1;
+  }
+  s = hts_getCharsetFromMeta(argv[0], strlen(argv[0]));
+  printf("%s\n", s != NULL ? s : "(none)");
+  freet(s);
+  return 0;
+}
+
+static int st_isutf8(httrackp *opt, int argc, char **argv) {
+  (void) opt;
+  if (argc < 1) {
+    fprintf(stderr, "isutf8: needs a string\n");
+    return 1;
+  }
+  printf("%d\n", hts_isStringUTF8(argv[0], strlen(argv[0])) ? 1 : 0);
+  return 0;
+}
+
 static int st_idna_encode(httrackp *opt, int argc, char **argv) {
   char *s;
 
@@ -1500,6 +1623,18 @@ static int st_cache_corrupt(httrackp *opt, int argc, char **argv) {
   return err;
 }
 
+static int st_cache_legacy(httrackp *opt, int argc, char **argv) {
+  int err;
+
+  if (argc < 1) {
+    fprintf(stderr, "cache-legacy: needs a directory\n");
+    return 1;
+  }
+  err = cache_legacy_refused_selftest(opt, argv[0]);
+  printf("cache-legacy: %s\n", err ? "FAIL" : "OK");
+  return err;
+}
+
 static int st_reconcile(httrackp *opt, int argc, char **argv) {
   int err;
 
@@ -1551,6 +1686,26 @@ static int st_cookies(httrackp *opt, int argc, char **argv) {
     err = 1;
   if (strstr(hdr, "junk") != NULL) // wrong-domain cookie leaked
     err = 1;
+#ifndef _WIN32
+  /* the jar holds live session cookies: cookie_save must keep it 0600 */
+  {
+    const char *jar = "st-cookies-jar.txt";
+    struct stat st;
+
+    (void) UNLINK(jar);
+    assertf(cookie_save(&cookie, jar) == 0);
+    assertf(stat(jar, &st) == 0);
+    assertf((st.st_mode & 07777) == HTS_PROTECT_FILE);
+    assertf(st.st_size > 0); /* mode-only checks would pass an empty jar */
+    /* a pre-existing world-readable jar must be tightened, not kept */
+    assertf(chmod(jar, 0644) == 0);
+    assertf(cookie_save(&cookie, jar) == 0);
+    assertf(stat(jar, &st) == 0);
+    assertf((st.st_mode & 07777) == HTS_PROTECT_FILE);
+    assertf(st.st_size > 0);
+    (void) UNLINK(jar);
+  }
+#endif
   printf("cookie-header: %s\n", err ? "FAIL" : "OK");
   if (err)
     printf("  got: %s\n", hdr);
@@ -2247,6 +2402,10 @@ static const struct selftest_entry {
     {"filtersize", "<size> <string> <filter>...",
      "size-aware filter verdict (negative size = unknown/scan time)",
      st_filtersize},
+    {"filtermime", "<mime> <filter>...",
+     "mime-type filter verdict (fa_strjoker type=1)", st_filtermime},
+    {"filtermemo", "[iterations]",
+     "memoized vs unmemoized matcher differential", st_filtermemo},
     {"filterdual", "<string1> <string2> <filter>...",
      "merged two-form filter verdict (fa_strjoker_dual)", st_filterdual},
     {"simplify", "<path>", "collapse ./ and ../ in a path", st_simplify},
@@ -2259,6 +2418,9 @@ static const struct selftest_entry {
     {"mime", "<filename>", "MIME type for a filename", st_mime},
     {"charset", "<charset> <string>",
      "convert a string to UTF-8 from a charset", st_charset},
+    {"metacharset", "<html>", "extract the <meta> charset from an HTML page",
+     st_metacharset},
+    {"isutf8", "<string>", "is the string valid UTF-8 (1/0)", st_isutf8},
     {"idna-encode", "<host>", "encode a hostname to IDNA/punycode",
      st_idna_encode},
     {"idna-decode", "<host>", "decode an IDNA/punycode hostname",
@@ -2294,6 +2456,8 @@ static const struct selftest_entry {
      st_cache_writefail},
     {"reconcile", "<dir>", "cache generation reconcile policy self-test",
      st_reconcile},
+    {"cache-legacy", "<dir>", "pre-3.31 legacy cache refusal self-test",
+     st_cache_legacy},
     {"cache-corrupt", "<dir>", "cache read-side corruption self-test",
      st_cache_corrupt},
     {"dns", "", "DNS resolver/cache self-test", st_dns},

tests/01_engine-charset.test

@@ -36,3 +36,64 @@ test "$(httrack -O /dev/null -#test=charset 'no-such-charset-xyz' 'café' 2>/dev
 # malformed UTF-8 (lone continuation byte, truncated lead byte) must not crash
 runs 'utf-8' $'\x80'
 runs 'utf-8' $'\xc3'
+
+# hts_getCharsetFromMeta: <meta> charset extraction, HTML5 and legacy forms.
+# -#test=metacharset <html> prints the charset, or "(none)".
+meta() {
+    test "$(httrack -O /dev/null -#test=metacharset "$1")" == "$2" || exit 1
+}
+
+# HTML5 form, quoting/spacing flavors
+meta '<meta charset="utf-8">' 'utf-8'
+meta "<meta charset='utf-8'>" 'utf-8'
+meta '<meta charset=utf-8>' 'utf-8'
+meta '<meta charset=utf-8 />' 'utf-8'
+meta '<meta charset=utf-8/>' 'utf-8'
+# charset labels are case-insensitive downstream; only require case-insensitive match
+test "$(httrack -O /dev/null -#test=metacharset '<META CHARSET=UTF-8>' | tr '[:upper:]' '[:lower:]')" == 'utf-8' || exit 1
+meta '<html><head><meta   charset = "utf-8"  ></head>' 'utf-8'
+
+# legacy http-equiv form, attributes in any order
+meta '<meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">' 'iso-8859-1'
+meta '<meta content="text/html; charset=gb2312" http-equiv="content-type">' 'gb2312'
+meta '<meta http-equiv=Content-Type content="text/html;charset=euc-jp">' 'euc-jp'
+meta '<meta http-equiv="Content-Type" content="text/html; charset = koi8-r ">' 'koi8-r'
+meta '<meta http-equiv="Content-Type" content="text/html; charset=utf-8;">' 'utf-8'
+meta '<meta http-equiv=content-type content=text/html;charset=utf-8 />' 'utf-8'
+meta '<meta http-equiv=content-type content=text/html;charset=utf-8>' 'utf-8'
+
+# first resolvable meta wins; unrelated metas are skipped
+meta '<meta name="viewport" content="width=1"><meta charset="koi8-r">' 'koi8-r'
+meta '<meta charset="utf-8"><meta http-equiv="Content-Type" content="text/html; charset=gb2312">' 'utf-8'
+
+# duplicate attribute: first wins, as in browsers
+meta '<meta http-equiv="Content-Type" content="text/html; charset=big5" http-equiv="refresh">' 'big5'
+
+# negatives: wrong tag/attribute, empty value
+meta '<p>x</p>' '(none)'
+meta '<meta name="desc" content="charset=bogus">' '(none)'
+meta '<meta http-equiv="refresh" content="0; url=x.html">' '(none)'
+meta '<body charset="utf-8">' '(none)'
+meta '<meta charset="">' '(none)'
+
+# hostile truncations: no crash, no match
+meta '<meta charset="utf-8' '(none)'
+meta '<meta charset=' '(none)'
+meta '<meta ' '(none)'
+meta '<meta' '(none)'
+
+# hts_isStringUTF8: valid-UTF-8 probe (guards the #180 link conversion).
+utf8() {
+    test "$(httrack -O /dev/null -#test=isutf8 "$1")" == "$2" || exit 1
+}
+utf8 'plain' '1'
+utf8 'café' '1'
+utf8 '统计' '1'
+utf8 "$(printf 'caf\xe9')" '0'              # latin-1 accent
+utf8 "$(printf '\xc3')" '0'                 # truncated lead byte
+utf8 "$(printf '\xa9')" '0'                 # lone continuation byte
+utf8 "$(printf '\xf0\x9f\x98\x80')" '1'     # 4-byte emoji
+utf8 "$(printf '\xc0\xaf')" '0'             # overlong (latin-1 "À¯" must stay convertible)
+utf8 "$(printf '\xed\xa0\x80')" '0'         # UTF-16 surrogate
+utf8 "$(printf '\xf4\x90\x80\x80')" '0'     # above U+10FFFF
+utf8 "$(printf '\xf8\x88\x80\x80\x80')" '0' # legacy 5-byte form

tests/01_engine-filter.test

@@ -131,3 +131,64 @@ match '*.aspx' 'page.aspx?y=2'
 # empty subject against a unique-match allow-all pattern (heap over-read regress)
 nomatch_empty '*(('
 nomatch_empty '**(('
+
+# near-miss negatives: word order, exact tails, range boundaries, class case
+nomatch 'foo*bar' 'foobaz'
+nomatch '*foo*bar' 'barfoo' # both words present, wrong order
+nomatch '*.gif' 'a.gifx'    # trailing junk after the extension
+nomatch '*.gif' 'a.gi'
+nomatch '*[b-y]' 'a'
+nomatch '*[b-y]' 'z'
+nomatch '*[a-z].txt' 'B.txt' # classes are case-sensitive...
+match 'FoO*' 'foobar'        # ...literals are not
+
+# a star may match an empty run, but never skip a required literal
+match '*[0-9].gif' '.gif'
+nomatch '*[0-9].gif' 'x.gif'
+
+# a descending range *[c-a] is an empty class, not c..a reversed
+match '*[c-a]x' 'x'
+nomatch '*[c-a]x' 'cx'
+
+# *[param] is all-or-nothing on '?': spans everything after one, else nothing
+match 'a*[param]' 'a?x=1'
+nomatch 'a*[param]' 'ab?x=1'
+
+# the latest matching rule wins, regardless of sign
+fsize 'verdict=forbidden size_flag=0' -1 a.gif '+*.gif' '-*.gif'
+fsize 'verdict=allowed size_flag=0' -1 a.gif '-*.gif' '+*.gif'
+
+# a failed size test sets size_flag but leaves the rule unmatched
+fsize 'verdict=unknown size_flag=1' 5 foo.jpg '-*.jpg*[>10]'
+
+# mime filters (-#test=filtermime = fa_strjoker type=1): only mime: rules
+# apply to a mime type, and mime: rules never apply to a URL
+fmime() {
+    local want="$1"
+    shift
+    test "$(httrack -O /dev/null -#test=filtermime "$@")" == "verdict=$want" || exit 1
+}
+fmime allowed 'image/png' '+mime:image/*'
+fmime unknown 'text/html' '+mime:image/*'
+fmime forbidden 'image/png' '+mime:image/*' '-mime:image/png' # last wins
+fmime allowed 'image/png' '-mime:image/png' '+mime:image/*'
+fmime unknown 'image/png' '+image/*'                               # URL rule: skipped
+fsize 'verdict=unknown size_flag=0' -1 'image/png' '+mime:image/*' # and vice versa
+
+# memoized vs no-memo differential over seeded random patterns and subjects,
+# matching and non-matching (the self-test asserts both polarities occurred)
+test "$(httrack -O /dev/null -#test=filtermemo)" == "filtermemo: 20000 cases OK" || exit 1
+
+# star-heavy non-match must stay polynomial (#501); 400 chars also exercises
+# the heap-allocated memo. timeout (where available) turns a hang into a FAIL.
+with_timeout() {
+    if command -v timeout >/dev/null; then timeout 60 "$@"; else "$@"; fi
+}
+pat=$(printf '*[a]%.0s' $(seq 1 12))b
+subj=$(printf 'a%.0s' $(seq 1 400))
+test "$(with_timeout httrack -O /dev/null -#test=filter "$pat" "$subj")" == "$subj does NOT match $pat" || exit 1
+
+# a later star's class dead-end must re-extend the earlier star, not give up
+# (guards against a single-backtrack-point matcher rewrite)
+match '*[aX]X*[a]Y' 'aXaXaY'
+nomatch '*[aX]X*[a]Y' 'aXbXaY'

tests/01_zlib-cache-legacy.test

@@ -0,0 +1,23 @@
+#!/bin/bash
+#
+# The pre-3.31 .dat/.ndx cache import was removed: cache_init must refuse the
+# legacy pair and leave the files untouched (httrack -#test=cache-legacy <dir>).
+
+set -eu
+
+dir=$(mktemp -d)
+trap 'rm -rf "$dir"' EXIT
+
+# the refusal errors land on stdout (no log file); pin them and the verdict
+out=$(httrack -#test=cache-legacy "$dir" 2>/dev/null)
+cnt=$(printf '%s\n' "$out" | grep -c 'no longer supported' || true)
+test "$cnt" = 2 || {
+    echo "expected 2 refusal messages, got $cnt" >&2
+    exit 1
+}
+out=$(printf '%s\n' "$out" | tail -n1)
+
+test "$out" = "cache-legacy: OK" || {
+    echo "expected 'cache-legacy: OK', got: $out" >&2
+    exit 1
+}

tests/11_crawl-international.test

@@ -22,9 +22,9 @@ bash crawl-test.sh \
     httrack http://ut.httrack.com/unicode-links/utf8.html
 
 bash crawl-test.sh \
-    --errors 4 --files 7 \
-    --found ut.httrack.com/unicode-links/café3860.html \
-    --found ut.httrack.com/unicode-links/café9fa8.html \
+    --errors 2 --files 9 \
+    --found ut.httrack.com/unicode-links/café3860.html \
+    --found ut.httrack.com/unicode-links/café9fa8.html \
     --found ut.httrack.com/unicode-links/café30f4.html \
     --found ut.httrack.com/unicode-links/café5e1f.html \
     --found ut.httrack.com/unicode-links/café7b30.html \

tests/41_local-utf8-link.test

@@ -0,0 +1,21 @@
+#!/bin/bash
+#
+# #180: raw non-ASCII hrefs must reach the wire UTF-8 percent-encoded whatever
+# the declared page charset; each PDF exists only at its exact UTF-8 path.
+
+set -euo pipefail
+: "${top_srcdir:=..}"
+
+bash "$top_srcdir/tests/local-crawl.sh" --errors 0 --rerun \
+    --found 'charset/header/统计大数据服务平台.pdf' \
+    --found 'charset/meta5/统计大数据服务平台.pdf' \
+    --found 'charset/metaeq/统计大数据服务平台.pdf' \
+    --found 'charset/none/统计大数据服务平台.pdf' \
+    --found 'charset/latin1hdr/统计大数据服务平台.pdf' \
+    --found 'charset/latin1real/café.pdf' \
+    --found 'charset/metalatin1/déjà.pdf' \
+    --found 'charset/latin1ovl/À¡x.pdf' \
+    --found 'charset/priority/nuée.pdf' \
+    --found 'charset/preenc/统计大数据服务平台.pdf' \
+    --found 'charset/bom/统计大数据服务平台.pdf' \
+    httrack 'BASEURL/charset/index.html'

tests/Makefile.am

@@ -68,6 +68,7 @@ TESTS = \
 	01_zlib-acceptencoding.test \
 	01_zlib-cache.test \
 	01_zlib-cache-corrupt.test \
+	01_zlib-cache-legacy.test \
 	01_zlib-cache-golden.test \
 	01_zlib-cache-writefail.test \
 	01_zlib-savename-cached.test \
@@ -108,6 +109,7 @@ TESTS = \
 	37_local-cache-outage.test \
 	38_local-update-304.test \
 	39_local-delayed-cancel.test \
-	40_local-why.test
+	40_local-why.test \
+	41_local-utf8-link.test
 
 CLEANFILES = check-network_sh.cache

tests/local-server.py

@@ -660,6 +660,97 @@ class Handler(SimpleHTTPRequestHandler):
     def route_intl_page(self):
         self.send_raw(b"<html><body>accented page</body></html>\n", "text/html")
 
+    # Raw non-ASCII href matrix (#180): each variant declares the page charset
+    # differently; the PDF exists only at its exact UTF-8 path, so a
+    # mis-decoded link 404s.
+    CHARSET_CJK = "统计大数据服务平台.pdf"
+    # variant -> (index Content-Type, <head> bytes, href bytes, pdf name)
+    CHARSET_VARIANTS = {
+        "header": ("text/html; charset=utf-8", b"", CHARSET_CJK.encode(), CHARSET_CJK),
+        "meta5": (
+            "text/html",
+            b'<meta charset="utf-8">',
+            CHARSET_CJK.encode(),
+            CHARSET_CJK,
+        ),
+        "metaeq": (
+            "text/html",
+            b'<meta http-equiv="Content-Type" content="text/html; charset=utf-8">',
+            CHARSET_CJK.encode(),
+            CHARSET_CJK,
+        ),
+        "none": ("text/html", b"", CHARSET_CJK.encode(), CHARSET_CJK),
+        "latin1hdr": (
+            "text/html; charset=iso-8859-1",
+            b"",
+            CHARSET_CJK.encode(),
+            CHARSET_CJK,
+        ),
+        # genuine latin-1 href: the charset conversion must still apply
+        "latin1real": (
+            "text/html; charset=iso-8859-1",
+            b"",
+            "café.pdf".encode("latin-1"),
+            "café.pdf",
+        ),
+        # latin-1 declared by META only: the meta parser is load-bearing
+        "metalatin1": (
+            "text/html",
+            b'<meta charset="iso-8859-1">',
+            "déjà.pdf".encode("latin-1"),
+            "déjà.pdf",
+        ),
+        # latin-1 bytes that form an overlong UTF-8 shape: strict validation
+        # must still convert them
+        "latin1ovl": (
+            "text/html; charset=iso-8859-1",
+            b"",
+            "À¡x.pdf".encode("latin-1"),
+            "À¡x.pdf",
+        ),
+        # header wins over meta: latin-1 href only resolves if iso-8859-1 is kept
+        "priority": (
+            "text/html; charset=iso-8859-1",
+            b'<meta charset="utf-8">',
+            "nuée.pdf".encode("latin-1"),
+            "nuée.pdf",
+        ),
+        "preenc": ("text/html", b"", quote(CHARSET_CJK).encode(), CHARSET_CJK),
+        "bom": ("text/html", b"", CHARSET_CJK.encode(), CHARSET_CJK),
+    }
+
+    def route_charset(self):
+        path = unquote(urlsplit(self.path).path)
+        parts = path.split("/")
+        if path == "/charset/index.html":
+            self.send_html(
+                "".join(
+                    '\t<a href="%s/index.html">%s</a>\n' % (v, v)
+                    for v in self.CHARSET_VARIANTS
+                )
+            )
+            return
+        if len(parts) == 4 and parts[2] in self.CHARSET_VARIANTS:
+            ctype, head, href, pdf = self.CHARSET_VARIANTS[parts[2]]
+            if parts[3] == "index.html":
+                body = (
+                    b"<html><head>"
+                    + head
+                    + b'</head><body><a href="'
+                    + href
+                    + b'">doc</a></body></html>'
+                )
+                if parts[2] == "bom":
+                    body = b"\xef\xbb\xbf" + body
+                self.send_raw(body, ctype)
+                return
+            if parts[3] == pdf:
+                self.send_raw(self.FAKE_PDF, "application/pdf")
+                return
+        self.send_response(404)
+        self.send_header("Content-Length", "0")
+        self.end_headers()
+
     # resume / 416 loop (#206): the first GET stalls after a prefix so the crawl
     # can be interrupted (partial + temp-ref); every later request is 416.
     RESUME_PREFIX = b"PARTIAL-" + b"x" * 4096  # flushed before the stall
@@ -1194,6 +1285,9 @@ class Handler(SimpleHTTPRequestHandler):
         if path.startswith("/big/"):
             self.route_big()
             return True
+        if path.startswith("/charset/"):
+            self.route_charset()
+            return True
         # Match percent-encoded paths (accented #157 route) by their decoded form.
         handler = self.ROUTES.get(path) or self.ROUTES.get(unquote(path))
         if handler is not None: