Arch Linux

Related  FS#72359 

I have to correct myself a bit: with SQLCIPHER_BUNDLED=1, upstream actually includes a vendored copy of sqlcipher (sigh..), and in that case it is redundant for the PKGBUILD to pull in sqlcipher, both as part of `depends` and `makedepends`.

SQLCIPHER_BUNDLED=1 doesn't have any effect here. Only preloading sqlcipher.so.0 works.

> Comment by Antonio Rojas (arojas) - Sunday, 21 May 2023, 09:09 GMT
> SQLCIPHER_BUNDLED=1 doesn't have any effect here. Only preloading sqlcipher.so.0 works.

I've just now created a completely fresh VM based on archlinux-2023.05.03-x86_64.iso, using the included archinstall script with the Desktop profile, choosing Xfce. I applied the attached patch to the element.io PKGBUILD (remove sqlcipher from depends and makedepends, change SQLCIPHER_STATIC=1 to SQLCIPHER_BUNDLED=1, nothing else), built and installed it, and Seshat/the search in E2E encrypted rooms feature works fine.

This is what I've done right after rebooting the VM for the first time:

$ sudo pacman -Syu git && git clone https://gitlab.archlinux.org/archlinux/packaging/packages/element.io.git && cd element.io
$ git log
commit fed620451d1930699c83c2aefdc32250ca1b5cf6 (HEAD -> main, tag: 1.11.31-2, origin/main, origin/HEAD)
<snip>
$ git am 0001-Fix-FS-78527.patch # patch is attached to this post for reproducibility
$ gpg --import keys/pgp/712BFBEE92DCA45252DB17D7C7BE97EFA179B100.asc
$ makepkg -sic
(go through build process)
$ element-desktop
(feature works)

The current tip commit does not preload sqlcipher.

Could you please take a look at this again?

(sorry that the formatting turned out funny, I don't know what formatting features flyspray supports and there doesn't seem to be a preview feature)

Please try building it in a clean chroot

> Comment by Antonio Rojas (arojas) - Monday, 22 May 2023, 11:27 GMT
> Please try building it in a clean chroot

Right, thanks for the heads-up. If you adjust the default packaging options in the PKGBUILD, then the Seshat feature works if the executable was built inside a clean chroot too.

This is because the devtools override the default makepkg configuration within the chroot (compare https://gitlab.archlinux.org/archlinux/devtools/-/blob/1.0.0/config/makepkg/x86_64.conf#L97 to `$ man 5 makepkg.conf`), which results in parts of the package being removed after the build, leaving element unable to access the library.

The attached revised patch to the PKGBUILD works when building inside a clean chroot.
I set `options=(libtool staticlib !lto)` here which may not be optimal; it may be possible to override less options while still leaving the Seshat feature working, but as building element.io takes a while I haven't looked further into this.

This what I did after rebooting the VM for the first time to confirm this:

$ sudo pacman -Syu devtools && git clone https://gitlab.archlinux.org/archlinux/packaging/packages/element.io.git && cd element.io
$ git log
commit fed620451d1930699c83c2aefdc32250ca1b5cf6 (HEAD -> main, tag: 1.11.31-2, origin/main, origin/HEAD)
<snip>
$ git am 0001-Fix-FS-78527-v2.patch
$ gpg --import keys/pgp/712BFBEE92DCA45252DB17D7C7BE97EFA179B100.asc
$ extra-x86_64-build
(go through build process)
$ sudo pacman -U element-web-1.11.31-3-x86_64.pkg.tar.zst element-desktop-1.11.31-3-x86_64.pkg.tar.zst
$ element-desktop
(feature works)

Although it doesn't seem to be documented in the wiki, I've picked up that Arch Linux policy is to discourage static linking (devtools overriding makepkg's default package options the way they do probably has a good reason, I surmise!), and I'm not sure to what extent exceptions are granted.
Now that upstream has decided to vendor sqlcipher, I'd prefer the static linking over a LD_PRELOAD + runtime dependency solution if the latter could work too (Bruno also wrote here https://gitlab.archlinux.org/archlinux/packaging/packages/element.io/-/commit/871a2319478de9f278863ccb883e20311743d240 that this caused conflicts with GTK?), but this is of course package maintainers' decision to make.

What if you drop the options and add CFLAGS+=' -ffat-lto-objects' as in the attached diff. This is working on the assumption the flag that fixed the build was !lto. Sorry I can not test this myself as I do not have a matrix account.

Running with `LD_PRELOAD=/usr/lib/libsqlcipher.so /usr/bin/element-desktop` fixes this, so might not be necessary to switch to bundled sqlcipher.

Thanks @6t8k and @loqs, lto was the issue.

> as building element.io takes a while
yeah, unfortunately our electron package is badly outdated so it builds its own bundled electron.