Archived document, last updated August 2020

This was a comprehensive assessment that Zack Weinberg and Sumana Harihareswara developed for Changeset Consulting to itemize what would ideally produce the best results for Autoconf 2.70.

Autoconf 2.70 was released in November of 2020, with some but not all of this work done. The text below has not been updated since August 2020.

Autoconf 2.70 release long TODO

• Find not-yet-merged patches
  • Develop list of "downstream" redistributors of Autoconf to check for desirable patches, starting with:
    • Debian
    • Fedora
    • Ubuntu
    • Arch
    • OpenEmbedded
    • Yocto
      https://lists.gnu.org/archive/html/autoconf/2020-03/msg00070.html
    • OpenBSD
    • FreeBSD
    • NetBSD
  • Inventory and copy patches from distributions' patched builds into Savannah (estimate: 5-10 patches per downstream, many duplicates) [this + above: about 2 days of work]
  • Inventory and migrate patches from mailing list into Savannah (estimate: 300-500 patches)
    • can be parallelized, might take a week or 2
• Prioritize, review, and merge patches (1 person-month + MongoDB high-priority work parallelizable)
  • Deduplicate patches and match to bug reports
    • estimate: 2-3 full-time weeks. parallelizable
  • Prioritize patches to merge, miminizing disruptive changes
    • Patches that can be merged as-is with low risk - merge during beta
    • Patches needing significant additional work - consult MongoDB for desirability
      • if low-priority: defer to 2.71
      • if high-priority: rework and merge
    • Bugs with no patch - consult MongoDB again, but defer if possible
    • estimate: < 1 week
  • Merge fixed bugs
    • estimate: 1 week if no problems discovered in the process
• Find regressions
  • Recruit testers
    • Maintainers of closely related projects (a few hours)
      • Automake
      • Libtool
      • Gnulib
      • Autoconf Macro Archive
    • Research & make public announcements of beta & any interim releases (total 1 day, spread throughout beta period)
      • LWN
      • Linux & BSD distro primary development lists
        • debian-devel
      • .....
    • Important autoconf-using projects (a few hours)
      • (see list under "run autoconf on these projects')
    • social media (3 hours)
    • podcasts/lightning talks online (1 day)
  • Run bundled testsuite on as many OS and CPU combinations as possible
    • GCC Compile Farm: (takes about a day to grind through these, Contractor B)
      • aarch64-unknown-linux-gnu
      • armhf-unknown-linux-gnu
      • i686-unknown-linux-gnu
      • mips64el-unknown-linux-gnu
      • mips64-unknown-linux-gnu
      • powerpc64el-unknown-linux-gnu
      • powerpc64-unknown-linux-gnu
      • powerpc-ibm-aix7.1.3.0
      • powerpc-ibm-aix7.2.0.0
      • sparc-sun-solaris2.10
      • sparc-sun-solaris2.11
      • sparc64-unknown-linux-gnu
      • x86_64-unknown-linux-gnu
        • Debian unstable
        • CentOS 5 (about the oldest Linux anyone still uses AFAIK).
      • x86_64-unknown-freebsd12.1
      • x86_64-unknown-openbsd6.6
    • Recruit other OS/CPU combinations (estimate: 4 hours)
      • Contact Nelson Beebe <beebe@math.utah.edu> for access to "zoo" of other Unixes
        • email sent 2020-03-10
          no response as of 2020-06-11
      • Contact OpenCSW <dam@opencsw.org> http://new.opencsw.org/standards/build_farm for access to Solaris 8 and 10
        • email sent 2020-03-10
          no response as of 2020-06-11
      • Per Bothner <per@bothner.com> offers to test on MacOS (Homebrew) and Windows (WSL).
      • "Nightstrike" says help with MinGW is available on IRC (http://irc.oftc.net/#mingw-w64).
  • Run Automake and Libtool test suites on all the above platforms, with the new autoconf (takes about a day, Contractor B)
  • Regenerate packages' configure script using the new autoconf, investigate what changes in `config.status` (1 day to run; Contractor B)
    • coreutils-8.31
    • emacs-26.3
    • gcc-9.2
    • glib-2.63.6
    • python-3.8.2
  • Do a complete rebuild of at least one Linux distribution with the new autoconf
    • Gentoo suggested as a source-based distribution and source of volunteer testing
    • Debian has some automation for this but the docs haven't been updated in a long time (see https://wiki.debian.org/qa.debian.org/ArchiveTesting).
    • estimate: up to one week of CPU time + bug triage (ideally, Contractor B)
  • Receive and triage problem reports from the community
    • 1-4 hours/week throughout the beta period, so, about 2-6 days of work, estimating 4 days for cost purposes
• Fix regressions in dev trunk found by above testing (parallelizable)
  • estimate: 1-6 person-weeks depending on how many bugs are found where we absolutely cannot defer to 2.71 or quickly find & revert the problematic patch (estimating 2 weeks for cost purposes)
  • Known problems in Autoconf proper
    • Fix `make distcheck` failures when /bin/sh is dash
      77: Forced re-exec with CONFIG_SHELL FAILED (m4sh.at:68)
      78: Configure re-execs self with CONFIG_SHELL FAILED (m4sh.at:132)
      • Actually a bashism somewhere in the testsuite driver script; still under investigation (1-2 days debugging + review turnaround time)
      • Appeared to be `make distcheck`-specific due to an error in the documentation
        documentation corrected in 2d06297bd550a0015f3ccedda9f43d7c616cfaf8
      • Exposed by a bug in _AS_DETECT_BETTER_SHELL
        caused by 2b59b6f8a79b8bf77e178ff4e5aa0ede433d39cf, fixed in 9894c7b5ba694820636f79ddec0c82e23af5e8e0
    • Patches needing review:
      • https://savannah.gnu.org/support/?110210 and https://savannah.gnu.org/support/?110211
        • https://lists.gnu.org/archive/html/autoconf-patches/2020-03/msg00012.html
        • https://lists.gnu.org/archive/html/autoconf-patches/2020-03/msg00013.html
        • https://lists.gnu.org/archive/html/autoconf-patches/2020-03/msg00014.html
  • Known problems exposed in other projects
    • Fallout from obsoletion of $as_echo
      • used by libtool/tests/testsuite.at
        patch submitted: https://lists.gnu.org/archive/html/libtool-patches/2020-03/msg00000.html reviewed but not committed
        • nudge maintainers (1h)
      • used by autoconf-archive/m4/ax_pthread.m4
        patch submitted, no response: https://savannah.gnu.org/patch/?9906
        • nudge maintainers (1h)
      • used by gnulib-common.m4
        reported to bug-gnulib: https://lists.gnu.org/archive/html/bug-gnulib/2020-03/msg00032.html discussion immediately went down a rabbit hole, no resolution
        • nudge maintainers (1h)
      • $as_echo(_n) used directly by various configure.ac's
        backward compatibility code added in 8a09003664733eb6bef031f56ef9cd575e403023
• Polishing
  • Go through documentation and code, find and correct all references to "future" events that have already happened
  • Review in-tree notes files (README*, BUGS, HACKING, INSTALL, TODO) and remove obsolete entries
  • Double-check that nothing has been left out of NEWS
• Cut the release
  • Bureaucratic operations - tag git, create tarballs, upload to ftp.gnu.org (a few hours)
  • Make announcements (a few hours)
• Post-release (27.1 work):
  • Feature: Per https://lists.gnu.org/archive/html/autoconf/2020-03/msg00021.html Debian intentionally configures /bin/dash without support for $LINENO to force configure scripts to use bash instead, due to bashisms in many configure scripts
    • have Debian do an archive rebuild with $LINENO turned back on and a mass bug filing
    • add to Autoconf an explicitly supported method for forcing ./configure to be a bash or zsh script
    • once that option exists, Autoconf should actively refuse to use bash or zsh unless that option has been set or it can't find any other vaguely POSIX compliant sh
  • Robust test suite:
    • _AS_DETECT_BETTER_SHELL needs more thorough testing; a catastrophic bug went unnoticed since 2013
    • Better way of detecting that "sh -n" isn't buggy
    • In general, all of the macros need more thorough automated tests