I think one quick workaround could be to include the 64-bit built Firefox part, too, in the bundling step and making sure we take the mar-tools from that one for generating the mar files. It's clumsy and I hope we can make the workaround even less so, but that's the best idea I can currently come up with.
However, what we actually want is to have both 32-bit and 64-bit mar-tools built when building the 32-bit Firefox code. That way we'd skip the inclusion and build of an additional 64-bit version of the firefox project if one just wants to build the 32-bit version and nothing else. We'd then extract the 64-bit mar-tools for update file generation in the 32-bit tor-browser build script while bundling the 32-bit mar-tools as usual.
Oh, and what we additionally want is to find this issue earlier (even though it should not happen that often). So, some check that .mar files actually got built when they are supposed to get built would be smart (I'll leave that here even though it's probably worth a new ticket).
So, the problem is we are using the 32bit mar-tools .zip file on a 64bit machine. Which does not work very well.
Hmm, I am not sure exactly why exactly it does not work, as in theory we should be able to run 32bit binaries in our 64bit container. So maybe this is something we can fix, after finding what is causing the issue.
In logs/tor-browser-linux-i686.log I can see some errors like:
Adding symlink add instructions to update manifestsAdding file and directory remove instructions from file 'removed-files'/var/tmp/tmp.on3obHmYW7/mar-tools/make_full_update.sh: line 170: /var/tmp/tmp.on3obHmYW7/mar-tools/mar: No such file or directorymv: cannot stat `/var/tmp/dist/tor-browser/tor-browser_zh-TW/Browser.work/output.mar': No such file or directoryFinished
Oh, and what we additionally want is to find this issue earlier (even though it should not happen that often). So, some check that .mar files actually got built when they are supposed to get built would be smart (I'll leave that here even though it's probably worth a new ticket).
Oh, and what we additionally want is to find this issue earlier (even though it should not happen that often). So, some check that .mar files actually got built when they are supposed to get built would be smart (I'll leave that here even though it's probably worth a new ticket).
Yeah, with the error that's true. I was just wondering whether we should safeguard against .mar files not being created even if no errors during the creation steps themselves are thrown (because maybe they are not even triggered or whatever). However, I am fine starting with #26907 (moved) first and then think more about it.