View all weekly reports

Reproducible Builds: Weekly report #7

Published: Jun 15, 2015.


What happened about the reproducible builds effort for this week:

Presentations

On June 7th, Reiner Herrmann presented the project at the Gulaschprogrammiernacht 15 in Karlsruhe, Germany. Video and audio recordings in German are available, and so are the slides in English.

Toolchain fixes

  • Joachim Breitner uploaded ghc/7.8.4-9 which uses a hash of the command line instead of the pid when calculating a “random” directory name.
  • Lunar uploaded mozilla-devscripts/0.42 which now properly sets the timezone. Patch by Reiner Herrmann.
  • Dmitry Shachnev uploaded python-qt4/4.11.4+dfsg-1 which now outputs the list of imported module in a stable order. The issue has been fixed upstream. Original patch by Reiner Herrmann.
  • Norbert Preining uploaded tex-common/6.00 which tries to ensure reproducible builds in files generated by dh_installtex.
  • Barry Warsaw uploaded wheel/0.24.0-2 which makes the output deterministic. Barry has submitted the fixes upstream based on patches by Reiner Herrman.

Daniel Kahn Gillmor’s report on help2man started a discussion with Brendan O’Dea and Ximin Luo about standardizing a common environment variable that would provide a replacement for an embedded build date. After various proposals and research by Ximin about date handling in several programming languages, the best solution seems to define SOURCE_DATE_EPOCH with a value suitable for gmtime(3).

W. Martin Borgert wondered if Sphinx could be changed in a way that would avoid having to tweak debian/rules in packages using it to produce HTML documentation.

Daniel Kahn Gillmor opened a new report about icont producing unreproducible binaries.

Packages fixed

The following 32 packages became reproducible due to changes in their build dependencies: agda, alex, c2hs, clutter-1.0, colorediffs- extension, cpphs, darcs- monitor, dispmua, haskell- curl, haskell- glfw, haskell- glib, haskell- gluraw, haskell- glut, haskell- gnutls, haskell- gsasl, haskell- hfuse, haskell-hledger- interest, haskell- hslua, haskell- hsqml, haskell- hssyck, haskell-libxml- sax, haskell- openglraw, haskell- readline, haskell- terminfo, haskell-x11, jarjar-maven- plugin, kxml2, libcgi-struct-xs- perl, libobject-id- perl, maven-docck- plugin, parboiled, pegdown.

The following packages became reproducible after getting fixed:

Some uploads fixed some reproducibility issues but not all of them:

Patches submitted which did not make their way to the archive yet:

reproducible.debian.net

A new variation to better notice when a package captures the environment has been introduced. (h01ger)

The test on Debian packages works by building the package twice in a short time frame. But sometimes, a mirror push can happen between the first and the second build, resulting in a package built in a different build environment. This situation is now properly detected and will run a third build automatically. (h01ger)

OpenWrt, the distribution specialized in embedded devices like small routers, is now being tested for reproducibility. The situation looks very good for their packages which seems mostly affected by timestamps in the tarball. System images will require more work on debbindiff to be better understood. (h01ger)

debbindiff development

Reiner Herrmann added support for decompling Java .class file and .ipk package files (used by OpenWrt). This is now available in version 22 released on 2015-06-14.

Documentation update

Stephen Kitt documented the new --insert-timestamp available since binutils- mingw-w64 version 6.2 available to insert a ready-made date in PE binaries built with mingw-w64.

Package reviews

195 obsolete reviews have been removed, 65 added and 126 updated this week.

New identified issues:

Misc.

Holger Levsen reported an issue with the locales-all package that Provides: locales but is actually missing some of the files provided by locales.

Coreboot upstream has been quick to react after the announcement of the tests set up the week before. Patrick Georgi has fixed all issues in a couple of days and all Coreboot images are now reproducible (without a payload). SeaBIOS is one of the most frequently used payload on PC hardware and can now be made reproducible too.

Paul Kocialkowski wrote to the mailing list asking for help on getting U-Boot tested for reproducibility.

Lunar had a chat with maintainers of Open Build Service to better understand the difference between their system and what we are doing for Debian.


View all weekly reports