Make the build reproducible

[lamby]

Whilst working on the Reproducible Builds effort, we noticed that strace could not be built reproducibly. This is due to the documentation encodes a timezone-varying date in the manpages.

(This was originally filed in Debian as #896016.)

[codecov-io]

Codecov Report

Merging #68 into master will decrease coverage by <.01%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master      #68      +/-   ##
==========================================
- Coverage   83.66%   83.65%   -0.01%     
==========================================
  Files         221      221              
  Lines       13711    13711              
  Branches      580      580              
==========================================
- Hits        11471    11470       -1     
- Misses       2107     2108       +1     
  Partials      133      133

Impacted Files	Coverage Δ
xmalloc.c	65.95% <0%> (-2.13%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ee2c065...a87b8ea. Read the comment docs.

[ldv-alt]

I'm sorry but your proposed commit just breaks the script it was supposed to fix:

+# Use SOURCE_DATE_EPOCH if it exists.
+# <https://reproducible-builds.org/specs/source-date-epoch/>
+[ -n "${SOURCE_DATE_EPOCH}" ] ||
+ date="@${SOURCE_DATE_EPOCH}"

Contrary to the comment, it ignores SOURCE_DATE_EPOCH if the latter exists, and it makes the script issue an error and print nothing if SOURCE_DATE_EPOCH does not exist.

Besides that, it was wrong to override DATE_FILE and "git log" with SOURCE_DATE_EPOCH.

[lamby]

Besides that, it was wrong to override DATE_FILE and "git log" with SOURCE_DATE_EPOCH.

Can you elaborate why? That's somewhat the point of this environment variable. :)

I can fix the technical aspects of this bug but we might as well agree first on the philosophy!

[esyr]

Because both of them are reproducible sources of the file date and make more sense than an arbitrary SOURCE_DATE_EPOCH.

I think that the only problems you actually had are lack of UTC option in date invocations and absence of LC_DATE reset, as .strace.1.in.date file used for identifying strace.1.in file date (the only one for which the script is used so far) is generated during build (and goes into the distribution tarball), and builds from git tree use git commit timestamps as a reference.

[ldv-alt]

SOURCE_DATE_EPOCH is not expected to override timestamps written in the source code.

[lamby]

Ah, I see our disconnect now.. so in Debian we try and rebuild everything and — as we don't have your .git tree at that point as we build from tarballs — we don't have a date.

Updated patch attached that prefers the git tree so the release behaviour is unchanged and also doesn't break if SOURCE_DATE_EPOCH is undefined.

:)

[ldv-alt]

Looks like you build from git snapshots (produced by git archive --format=tar) instead of released tarballs (produced by make-dist script), otherwise SOURCE_DATE_EPOCH wouldn't make any difference.

Anyway, please have a look at the current master branch, I suppose it addresses the issue.

[lamby]

(Commented on that commit)