Hi, Publican 3.1.5 is now released, since we got distracted between
3.1.0 and now I'm going to include the intermediate releases in this list.
NOTE 1. Some of these are recent bugs and some are long standing bugs
that have not been brought to our attention before. One includes a patch
for an upstream module that system packagers will need to carry upstream
[1].
NOTE 2. I had to rework the spec file for wkhtmltoipdf-qt and
wkhtmltopdf to stop them interfering with the system Qt libraries. Rudi
or I will post updated packages soon for people suffering from Qt
packages (like Kate) blowing up.
3.1.1 Thu Feb 07 2013
- Fix web site CSS for admonitions breaking layout for books built on
P3.0. BZ #908539
3.1.2 Tue Feb 19 2013
- Fix tests failing when publican not installed. BZ #908956
- Fix broken mr-IN/Conventions.po. BZ #908956
- Fix footnote link unclickable. BZ #909006
- Fix missing translations for common files. BZ #908976
- Fix using edition for version on cover pages. BZ #912180
- Fix nested entities causing XML::TreeBuilder to fail. BZ #912187
- Fix for cover pages causing page number overrun. BZ #912967
3.1.3 Fri Feb 22 2013
- Fix add_revision breaking XML parser. BZ #912985
- Stronger fix for cover pages causing page number overrun. BZ #912967
- Fix CSS for article front page subtitle. BZ #913016
3.1.4 Tue Mar 12 2013
- Fix entities in Book_Info braking build. BZ #917898
- add translations of "Revision History". BZ #918365
- Fix TOC title not translated in PDF. BZ #918365
- Fix translated strings with parameters. BZ #891166
- update translations
- add it-IT translation of PUG via <[email protected]> BZ #797515
3.1.5 Mon Mar 18 2013
- Fix translated PDF encode issue when build from packaged books. BZ
#922618
https://fedorahosted.org/releases/p/u/publican/Publican-v3.1.5.tar.gz
https://fedorahosted.org/releases/p/u/publican/publican.spec
https://fedorahosted.org/releases/p/u/publican/publican-3.1.5-0.el6.src.rpm
I promised a report on the 2013 tour of Hell, I[ve shortened it somewhat
and hope it's sufficient.
Up until 3.1.1 was released we had no problems building PDFs with
wkhtmnltopdf. I'd built hundreds of PDFs with it, never a single issue
with its stability. After releasing 3.1.1 we started to get massive
failure rates in building PDFs, in excess of 80%. An assert was getting
triggered that the page numbers exceeded the expected number of pages.
The effect of this on RHEL was an order of magnitude greater than the
effect on Fedora; on Fedora 17 we where getting 1-2% failure when
building a book in batches of 100. i.e. building the same PDF calling
wkhtmltopdf directly in the CLI in a for loop.
After a great deal of poking around it became apparent that the event
driven mechanism being used was creating sub-events that where
completing after the main event and that the book was starting to print
before the cover pages had finished loading.
3.1.2 contained a fix where we reduced the number of cover pages from 2
to 1. This seemed to have a significant affect on the ASSERT triggering.
However the effect in production was still too high for production use.
3.1.3 contained a fix where we significantly reduced the size and
complexity of the CSS being loaded in the cover page, and this combined
with a single cover page has led to us not seeing an ASSERTS since
making this change.
Once 3.1.3 went live more people started building books and several
issues that may have been detected much earlier where raised. Some of
them were very serious and required immediate attention. Some other have
probably been around for a long time but were also considered urgent.
3.1.4 contained those fixes.
3.1.5 has a fix for a UTF8 issue.
All up it's been quite a difficult journey and I apologise to all those
affected by this bumpy ride. I'm not sure how we can mitigate the PDF
ASSET failure. It came completely out of the blue and we still have no
idea what changed to start it triggering.
We have collected a broad selection of documents to test. We will test
all of these books every release, and will ensure the full test suite is
run on every build.
In the end, as much as we automate testing, a lot of these issues can
only be picked up by people who know what things should look like
actually looking at them, before we ship. Our automation will be based
on this. We will build your books, we will host the output, but we will
always need you to look at it and tell us if anything is wrong. This is
particularly true for translated content, we just can't tell if it's
right or wrong.
I'd like to give a shout out to Raphaël Hertzog who packages Publican
for Debian and who gives me a lot of good feedback on how to make the
packaging experience better. I really appreciate your time and efforts
mate, I'll buy you a beer one day if we ever get to meet.
Cheers, Jeff.
1: Bug 891166 appears to be a long standing bug where strings with
parameters would not get translated. I had to patch
Locale::Maketext::Gettext to convert between gettext and maketext
formats. The patch probably isn't robust. I'll be opening an upstream
bug to get it carried. Packagers need to determine if it's worth the
risk carrying it in their distributions.
--
Jeff Fearn <[email protected]>
Senior Software Engineer
Infrastructure Engineering & Development (AEU)
Red Hat Asia Pacific Pty Ltd
GPG: 0x0357E8F0
diff -ru Locale-Maketext-Gettext-1.27/lib/Locale/Maketext/Gettext.pm Locale-Maketext-Gettext-1.27-patched/lib/Locale/Maketext/Gettext.pm
--- Locale-Maketext-Gettext-1.27/lib/Locale/Maketext/Gettext.pm 2009-04-28 03:46:23.000000000 +1000
+++ Locale-Maketext-Gettext-1.27-patched/lib/Locale/Maketext/Gettext.pm 2013-03-08 12:31:37.166997436 +1000
@@ -354,6 +354,10 @@
# Translated message
$strt = substr($content, $off, $len);
+ # Convert gettext params %1 to maketext params [_1]
+ $stro =~ s/\%(\d+)/\[_$1\]/g;
+ $strt =~ s/\%(\d+)/\[_$1\]/g;
+
# Hash it
$_{$stro} = $strt;
}
_______________________________________________
publican-list mailing list
[email protected]
https://www.redhat.com/mailman/listinfo/publican-list
Wiki: https://fedorahosted.org/publican
