[RFC] set timestamps in a package only for files newer than commit date

[tornaria]

The hook pre-pkg/90-set-timestamps.sh touches all files in a pkg to set the mtimes to the commit date. When a package installs files verbatim from its source, it might make sense to keep the original timestamp. This could be acomplished by using find predicate -newermt as in:

--- a/common/hooks/pre-pkg/90-set-timestamps.sh
+++ b/common/hooks/pre-pkg/90-set-timestamps.sh
@@ -5,6 +5,6 @@ hook() {
        # If SOURCE_DATE_EPOCH is set, set mtimes to that timestamp.
        if [ -n "$SOURCE_DATE_EPOCH" ]; then
                msg_normal "$pkgver: setting mtimes to %s\n" "$(date --date "@$SOURCE_DATE_EPOCH")"
-               find $PKGDESTDIR -print0 | xargs -0 touch -h --date "@$SOURCE_DATE_EPOCH"
+               find $PKGDESTDIR -newermt "@$SOURCE_DATE_EPOCH" -print0 | xargs -0 touch -h --date "@$SOURCE_DATE_EPOCH"
        fi
 }

I am thinking on a package like pari-elldata which ships only data files from its source. The alternative I found is to set XBPS_USE_BUILD_MTIME=no in the template, but (a) that is only good for this pkg because I know all the files in the package have a deterministic timestamp and (b) xlint complains when using this variable.

Rationale: the timestamp of the original files in the source is more canonical than the commit date.

[Chocimier]

Using commit date is not for canonicality but reproducibility. Time should be adjusted even when build machine clock is set eariler than commit date.

Are there any problems with pari-elldata files having build date?

[tornaria]

Using commit date is not for canonicality but reproducibility. Time should be adjusted even when build machine clock is set eariler than commit date.

I wouldn't know how to avoid that in a simple way, is that a normal issue? Maybe the current date can be compared with the commit date and fail if it's older?

Are there any problems with pari-elldata files having build date?

Not really. It just seemed a good idea to ship unmodified files with their original dates. Maybe I'm overthinking it. For the packages in question (PR #32641) I set XBPS_USE_BUILD_MTIME=no. Nothing breaks without this.

[ericonr]

XBPS_USE_BUILD_MTIME=no

I don't think it's good style to poke at such bits. It's not xbps-src "API".

[tornaria]

XBPS_USE_BUILD_MTIME=no

I don't think it's good style to poke at such bits. It's not xbps-src "API".

Ok, I removed that in #32641, and I'm closing this now. Thanks for your feedback.