If we read in a line with \0 at the beginning, blen will be 0. However,
we then try to index our copy of the buffer with
s->lines[s->linecount][blen-1], we'll read (and potentially write if the
data happens to be 0x0A) outside of strdup's allocated memory, and may
crash.
Fix this by just rejecting lines with a leading \0. Lines with nulls
embedded in other places don't invoke similar behaviour, since the
length is still >0.
Otherwise, on reloading sent would segfault. Move the check-logic to
load() rather than doing it in main() to prevent other misuses of it in
the future.
Previously when covering the sent window with an image slide selected and
exposing it again, the image was lost. This change always redraws the image and
also removes the now redundant bit flag in the state enum.
die() calls exit(), but actually at this point the program could be in a
state not allowing this. It's better to call _exit to stop the libc from
calling any handlers that might fail at this moment anyway.
Now, before you call your lawyers, here is some explanation:
The MIT/X Consortium-License and ISC-License are equivalent
according to the Berne-convention. This means that a court
of law presented with either licenses will act the same.
The advantage of the ISC license is that it has a simpler wording
and is recommended by the OpenBSD-project for new projects[0].
And as we at suckless.org love removing cruft, this also includes
removing legal cruft wording that is not needed any more.
[0]: http://www.openbsd.org/policy.html
This fixes a bug where the image would not be loaded if the user setup a
keybinding which advances multiple slides at once. Then the preloading for the
previous/next slide in advance() would not catch that case and instead of an
image the text version (`@filename`) would be rendered on the slide.
On Sat, Jun 04, 2016 at 05:28:31PM +0200, Markus Teich wrote:
> Hiltjo Posthuma wrote:
> > previously an image file would be opened but only ffread when advancing to
> > the slide, but when the slide was not used it gave an error:
> >
> > /usr/local/bin/2ff: failed to convert image/png
>
> Heyho Hiltjo,
>
> thanks for the patch. Unfortunately it does not work if the first slide contains
> an image (ffopen is only called in advance). I think it would be good to merge
> ffopen and ffread instead into a single function ffload. This makes the `LOADED`
> state clearer and also enforces that the fd is closed in the same function where
> it is opened. This ffload function should then be called in advance() replacing
> the ffread() calls if the image is not loaded yet and once in load() for the
> first slide if it is an image.
>
Ah yes, sorry for the oversight.
> If you want to take this new approach, go for it, otherwise I'll look into it
> myself.
>
I have attached a patch that does this, I hope you'll like it.
Also I'm not sure if we need the below code in advance(), I have removed it in
the patch:
if (slidecount > idx + 1 && slides[idx + 1].img)
ffread(slides[idx + 1].img);
if (0 < idx && slides[idx - 1].img)
ffread(slides[idx - 1].img);
That seems to preload the next and previous slide image right? A minor issue I
notice also is that images seem to flicker, it uses XPutImage directly to
xw.win. Maybe it can be replaced with a backbuffer then XCopyArea? What do you
think?
In advance() it should also not always be needed to rescale the image.
--
Kind regards,
Hiltjo
From 97bebdcab4003f9acdfdd4bdf424449299ffd61d Mon Sep 17 00:00:00 2001
From: Hiltjo Posthuma <hiltjo@codemadness.org>
Date: Sat, 4 Jun 2016 21:34:25 +0200
Subject: [PATCH] merge ffread and ffopen into one function
previously an image file would be opened but only ffread when advancing to
the slide, but when the slide was not used it gave an error:
/usr/local/bin/2ff: failed to convert image/png
this changes it to load the image on-demand once and "cache" it.
The child thread was created because execlp will not return if successful.
The eprintf was placed after the call to execlp to catch any error, but the
child continued to return a (closed) fds[0], resulting in a second slideshow
being run by the child.
This commit fixes the issue by killing the child.
