Am 20.09.20 um 00:35 schrieb Junio C Hamano: > René Scharfe <l.s.r@xxxxxx> writes: > >> Am 19.09.20 um 03:41 schrieb Junio C Hamano: >>> * jc/dist-tarball-tweak (2020-09-09) 1 commit >>> (merged to 'next' on 2020-09-10 at 36cbe7ee9e) >>> + Makefile: allow extra tweaking of distribution tarball >>> >>> Allow maintainers to tweak $(TAR) invocations done while making >>> distribution tarballs. >> >> Just noticed this one. It reminds me of an alternative solution for >> archives containing both tracked and untracked files gathering dust on >> my disk because I didn't see much demand. It goes the other way and >> gives untracked files the same meta data as tracked ones. Interested? >> >> archive: read short blobs in archive.c::write_archive_entry() >> archive: add --add-file >> Makefile: use git-archive --add-file > > Oooh, so is the idea that we do not have to use $(TAR) to append > untracked ones to "git archive" output, etc.? If we can get rid of > all $(TAR) invocations from the build procedure, that would be an > interesting addition. The remaining ones are used in the targets install and install-man-perl for copying po files, Perl libraries and manpages; in artifacts-tar for archiving binaries etc. and in dist-doc for archiving manpages and HTML files. git archive can only solve half of the copy use case at most, so $(TAR) will still be used -- unless it's replaced by cp+chmod or something else. Not sure how to support artifacts-tar. Allowing git-archive calls without commit or tree might be a start. Letting --add-file record the full relative path of an untracked entry instead of its basename would help, perhaps counterbalanced with an additional --strip option similar to the one of patch(1). dist-doc tars up directories. This would require adding support for reading directories to --add-file or adding --add-directory. René
