Make archives reproducible

[stertooy]

As suggested in #50.

The three supported formats, .tar.gz, .tar.bz2 and .zip should now always be reproducible archives. This is also tested by running this action twice, and then comparing the sha256 checksums of both archives.

Unfortunately, I had to add an extra input to be able to properly test this. I recommend leaving it undocumented, since it server no purpose for end users.

I've set SOURCE_DATE_EPOCH to the package release date as specified inPackageInfo.g.

[stertooy]

Weird thing I noticed: the sha256sum of ActionTestOldDocPackage is not the same between multiple runs. It appears this is because a .dvi file is created during the creation of the manual, but this file is not removed afterwards.

[limakzi]

@stertooy Isn't it like DVI file contains timestamp in a comment?
And more-over, documentation might have timestamp too?

[stertooy]

@limakzi It's very likely that a creation date timestamp somewhere in the .dvi or in its metadata causes this, yes. But the .dvi file shouldn't be in the release archives anyway, since it's only used as an intermediate step in producing the documentation.

This doesn't happen in the PDF's because the build-pkg-docs action sets SOURCE_DATE_EPOCH to 0.

[limakzi]

@stertooy I think we could replace affix with two different timestamps, for example?

[stertooy]

The idea is not to have any affix at all for end users, to keep things as simple as possible.

[limakzi]

@stertooy Why not?

Suggested change

	run: echo "${{ inputs.body-text }}" > $RUNNER_TEMP/Release_text_for_${PKGNAME}_${VERSION}${{ inputs.affix }}.md
	run: echo "${{ inputs.body-text }}" > $RUNNER_TEMP/Release_text_for_${PKGNAME}_${VERSION}_${{ inputs.affix }}.md

[stertooy]

Again, this is because the affix input is expected to be empty, except during the CI tests in this repo. The CI script will put either "-1" or "-2" in the affix input, so an extra underscore isn't needed I think.