Re: --with-cairo Emacs server crash and fix

[Liam Quinlan]

Alright cool.  I'll have a look at your changes for comparison.  Thanks for the explanation.

On Tue, Aug 13, 2019, 11:00 Eli Zaretskii, <address@hidden> wrote:

> From: Liam Quinlan <address@hidden>
> Date: Mon, 12 Aug 2019 22:52:31 -0400
> Cc: address@hidden
>
> Update: i think I've found the actual proper solution.  Alternative patch on offer below; this time by checking for
> dynamic bitmaps already in place after finishing the static ones in x_cr_init_fringe.  Note that that function also
> defines w32_init_fringe if HAVE_NTGUI, so windows is covered.

Yes, the idea is correct, but the implementation needed some tweaking,
in particular due to w32 doing things a bit differently during
startup.  See commit c90975f on the master branch.

> Trying to be minimal so this example doesn't iterate to over fringe_bitmaps looking for sparse elements, just
> begin checking fringe_bitmaps for existing elements at the index following the standard bitmaps leaves off and
> continue until the first null.  I'm not sure if it's too bold to trust that holes haven't had a chance to arise though.
> (perhaps if the server is initially used in console mode and packages get unloaded before a gui client
> connects...?).

No, we cannot assume there are no holes, you can clearly see in
destroy_fringe_bitmap that it leaves holes if the destroyed bitmap is
not the last one in the array.  So we need to initialize all the
user-defined bitmaps whose pointer is not NULL.

Btw, there's a simple way of reproducing the problem without any init
files and extra packages:

  emacs -Q -daemon
  emacsclient SOME-VERSIONED-FILE-UNDER-GIT
  M-x set variable RET diff-font-lock-prettify RET t RET
  [modify the file and then save it]
  C-x v =

Before the change I pushed the above would fail to display the vc-diff
bitmaps on the fringe (I presume the Cairo build will crash).  After
the change, the bitmaps are displayed correctly.  This happens because
diff-mode is loaded before the GUI frame is created, and thus the
fringe bitmaps defined by diff-mode during loading are not completely
initialized.

Thanks.