On Wed, May 4, 2022 at 10:35 PM Junio C Hamano <gitster@xxxxxxxxx> wrote: > Allowing translators to pick a random point in the history and run > the tool to generate .pot file locally creates that problem for us. > It also means that various .po files would correspond to slightly > different versions of the source tree. As Peter said [^1], even keeping "po/git.pot" unremoved, we can always regenerate the POT file locally instead of committing it, to update the PO file. We can add new checkpoints in l10n CI pipeline to check the baseline for each PO file. [^1]: https://lore.kernel.org/git/cb74f3b-c2e9-947f-8f89-f51e79b17825@xxxxxxxxxxxxxxxx/ > The "only the coordinator updates .pot, everybody works off of that > file" was one way to ensure that you do not have to worry about these > two problems. In the reply to Peter [^2], I demonstrated the effect of having a POT/PO file with or without location line numbers on repository size. We can mitigate the growth of the "po/" directory by removing "po/git.pot" and committing the PO files without the location line numbers. [^2]: https://lore.kernel.org/git/20220504124121.12683-1-worldhello.net@xxxxxxxxx/ > I am not sure what your plans are to make sure everybody works off > of the same version, though. Even if you scrape the source tree for > source files that may not be relevant to your build, if you do not > start off of the same version, your set of sources may be a bit off > compared to the other translators'. We can use "msgcmp" to compare PO files from l10n contributors with newly generated POT files and give suggestions in the l10n CI pipeline. -- Jiang Xin
