On 6/9/20 6:25 pm, Andreas Enge wrote: > (cc-ing this and only this bug, the other one seems to have diverged towards > css and typescript) > > On Fri, Sep 04, 2020 at 07:53:02PM +1000, Brendan Tildesley wrote: >>> I did not realise there was already an open ticket for updating calibre, >>> thanks for pointing it out. Indeed I do not think we need to wrap all >>> programs. I tried out the programs in ...calibre-4.18.0/bin, and only these >>> two fail with an error message: >>> ebook-edit >>> ebook-viewer >>> I did not try the different calibre-*; calibre itself starts. Then it can call >>> ebook-viewer, and I do not know the mechanism, so it might just call the >>> wrapped binary ebook-viewer, or it might need wrapping itself because of using >>> internal python mechanics. I would give it a try and not wrap it in the first >>> place. >>> >>> All "binaries" are already wrappend with PYTHONPATH, so there is probably some >>> mechanic to avoid double wrapping (which your patch may already address). >> Since it uses (replace 'wrap..., it won't run the old wrap phase any more. >> Also I don't think it matters much that the other variables also have the Qt >> variable wrapped, perhaps it is more correct anyway. Especially since the >> wrap script uses '=, which wrap-program interprets as overwriting the >> variable completely, so applying it twice won't make a difference anyway. > Well, I think we should not wrap more than absolutely necessary. And it would > even be easier to write (no need to look for all the binaries, just use these > two names). Would you like to create a patch maybe for the current calibre, > and handle the update following the other bug? > > Andreas > It's actually a bit more complicated because every other executable still requires PYTHONPATH wrapping anyway as python-build-system does it. I just tested creating such a wrap phase that only sets QTWEBENGINEPROCESS_PATH for ebook-viewer and ebook-edit. It works when they are opened directly, but when they are opened via the calibre interface, they fail to find qtwebengine. so it does seem that calibre it's self needs that variable set anyway. I've attached a patching fixing this one issue. For those I've emailed coming from the Issue I created for updating Calibre, since it looks like the update will be delayed until Mathjax 3 is packaged properly, would you mind just reviewing this fix to get it through for the current Calibre. I think Ricardo was having a go at packaging swc to transpile typescript which is needed for mathjax, but it looks like quite a challenge.