vcswatch reports that
this package seems to have new commits in its VCS but has
not yet updated debian/changelog. You should consider updating
the Debian changelog and uploading this new version into the archive.
Here are the relevant commit logs:
commit 3cbe7b6d9485b3db27adc37715728aaa5602cb58
Author: Xavier Roche <roche@httrack.com>
Date: Thu Aug 27 16:57:54 2026 +0200
A Romanian user is shown the English translators (#1454)
lang/Romanian.txt has never carried a value for the About blurb, so the
engine falls back to English and tells a Romanian user that Xavier Roche
and Robert Lagadec translated it. Confirmed by serving
/server/about.html at lang 25 before and after.
The value added here is the English text the fallback already renders,
with only the credit line changed to name Alin Gheorghe Miron, who is
Romanian's LANGUAGE_AUTHOR. Nothing a translator wrote is overwritten,
because there was nothing there. The body stays English until a Romanian
speaker translates it, which is what a reader sees today anyway.
Romanian's empty count in the untranslated pin drops from 5 to 4.
Emptying the value again, dropping the credit line, or crediting another
catalog's translator each fail a test.
Found while checking the fallout of #1453 against the other front ends.
---------
Signed-off-by: Xavier Roche <xroche@gmail.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
commit 4055ba5af2366181a7969282459d03f49c0e19b1
Author: Xavier Roche <roche@httrack.com>
Date: Thu Aug 27 14:30:20 2026 +0200
One name stood for every translator in the About box (#1453)
Every catalog's About box thanked Robert Lagadec for translation tips,
beside a copyright stamped 1998-2003 and a development line that still
described a single Windows interface. Twenty-five translators had
already replaced that line with their own credit. Four never did, and
Uzbek ended up thanking the Russian translator.
The English msgid loses the name and the year range. A year stamped into
a catalog string is what #1165 took out of the footers, and nothing here
would regenerate it. Translated values stay with their translators,
except in the five catalogs whose credit named the wrong person: there
only the name changes, and the wording around it is theirs. The credit
now comes from each catalog's LANGUAGE_AUTHOR, which the About page
renders, so every language names its own translators without anyone
editing prose they cannot read.
AUTHORS, greetings.txt and html/contact.html now carry one roster: the
translators the catalogs name, everyone git has recorded since, and the
older credits those lists already held. tests/372_credits.test holds the
three files to that roster by identity, and matches names as whole words
against the catalog's own row, so a new translator has to reach AUTHORS
under the right language.
The About block is wrapped in `${do:output-mode:html}`, which escapes
catalog text and keeps its newlines. That was already the server's idiom
for a `<pre>`, and it covers `${LANG_K1}`, which was interpolated raw
before.
One thing found on the way: tests/httpclient.py decoded every reply as
latin-1, so a test asserting non-ASCII text off a WebHTTrack page was
comparing re-encoded bytes.
WinHTTrack renders the same LANG_K1 and has no LANGUAGE_AUTHOR line, so
its About box shows the blurb without one until it picks the field up.
---------
Signed-off-by: Xavier Roche <xroche@gmail.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
commit c32299b5d4fbfb4d23b45d2141848feee800d89a
Author: Xavier Roche <roche@httrack.com>
Date: Thu Aug 27 10:06:55 2026 +0200
warc-offset fails on a filesystem that cannot hold a 1TB offset (#1451)
The probe seeks a file to 1TB-1 to prove the CDXJ record offset stays
64-bit past the 2GB `long` ceiling. GNU/Hurd's ext2fs caps file size
well below that, so `fseeko()` returns `EFBIG` and
`351_engine-warc-cache-io` reds on the [Debian
buildd](https://buildd.debian.org/status/package.php?p=httrack)'s
hurd-i386 leg. `01_engine-fsize` already skips on that host for the same
cap.
Offsets the filesystem refuses are now skipped and reported to stderr,
but only for `EFBIG`. The selftest also counts how many offsets past 2GB
it measured and fails when that count is zero, so a host that refuses
all of them cannot pass having proved nothing.
A new `fseeko()` interposer stands in for a filesystem with a low
ceiling. A cap that hides only the 1TB case still passes; a cap below
2GB has to fail on the tally; a refusal that is not `EFBIG` has to fail
too. Each of the three kills a mutant of the fix.
---------
Signed-off-by: Xavier Roche <roche@httrack.com>
Co-authored-by: Claude <noreply@anthropic.com>
commit fedaa90d57e96f8cacebff2efa7421d6274a0875
Author: Xavier Roche <roche@httrack.com>
Date: Thu Aug 27 08:50:28 2026 +0200
341 builds a near-miss host the server accepts by design (#1452)
`341_webhttrack-authority` truncates this host's own name to prove the
Host match is whole-name rather than a prefix. That is only a near-miss
when what is left is still a DNS name: `host_is_address()` accepts any
all-digits-and-dots authority as whoever opened the socket, so a
hostname like `12764103200x` truncates to something the panel answers at
by design. A uutils CI runner drew that hostname and the leg went red on
correct behaviour.
The name itself already goes through that filter before being used. This
reuses it for the truncation. Verified against the live server:
`12764103200` and `1.2.3.4` are accepted, while `rowie` (a real
truncation of this box's `rowie3`) is still refused, so the assertion
keeps its teeth on an ordinary host.
Signed-off-by: Xavier Roche <roche@httrack.com>
Co-authored-by: Claude <noreply@anthropic.com>
commit 4edfd40f259ea42b80ee2021e597726e0746f5d2
Author: Xavier Roche <roche@httrack.com>
Date: Thu Aug 27 08:39:37 2026 +0200
The crash report prints every signal number backwards (#1450)
The reversal loop in `print_num()` runs `j` across the whole range, so
every pair is swapped twice and the digits stay least-significant first.
Only signal numbers ever reach it, and on x86 those are 4, 6, 7 and 11,
all single digits or a palindrome, so nothing ever showed. hppa's SIGBUS
is 10, and the report there says `Caught signal 01`. That is what turns
`183_altstack-worker` red on the [Debian
buildd](https://buildd.debian.org/status/package.php?p=httrack).
The function moves to `htsbacktrace.c` beside the other signal-safe
writer so a test can link it, the way test 218 already links the crash
printer. Two latent problems in the same few lines go with it: negating
in signed `int` is UB for `INT_MIN`, and the returned length left out
the minus sign.
`372_crash-signal-number` grades the rendering against printf's, over
values that include 10, both digit parities and `INT_MIN`, into a
poisoned buffer so a stray terminator shows. It builds the probe with
the tree's own `CFLAGS`, so the sanitize leg catches the `INT_MIN` case
that no value comparison can see.
---------
Signed-off-by: Xavier Roche <roche@httrack.com>
Co-authored-by: Claude <noreply@anthropic.com>
![[bug history graph]](https://qa.debian.org/data/bts/graphs/h/httrack.png){kind=link}