I tried to keep the changes to existing tests as minimal as
possible. There were a few exceptions, though:
* test_update_ignorecompat.js was completely broken. I couldn't
figure out why it was suddenly failing after I changed it to use
`add_test`, and it turned out that it had been failing all along,
but in a way that the harness didn't pick up.
* I changed most of the `do_throw` in update callbacks to `ok(false`
because it took me about an hour to figure out where the test was
failing when I hit one of them.
* I made some changes to sync `test_update.js` and `test_update_ignorecompat.js`
where one appeared to have been changed without updating the
other.
* I made `promiseFindAddonUpdates` a bit more generic, because I was
planning to convert most of `test_update.js` to use it, rather
than nested callbacks. I changed my mind a quarter of the way
through, but decided to keep the changes, since they'll probably
be useful elsewhere.
--HG--
extra : commitid : 84oLUw4ZPOg
extra : rebase_source : 2bd6c921c6b677e4d487d0ee9c33b1130163bc39
Looking at a preference file read with strace typically looks like:
open("...", O_RDONLY) = X
...
read(X, "...", SIZE) = SIZE
read(X, "...", SIZE) = 0
...
There's no reason to call Read() and make another syscall to determine
there's no data left for reading. We can keep track of how much we've
read at minimal cost and thus determine for ourselves when we are done.
Calling nsIInputStream::Available on nsIFileInputStreams is relatively
expensive, as it requires three system calls: one to determine the
current file position, one to seek to the end file position, and one to
rewind to the previous file position.
We can do better by asking the file for its size directly, prior to
opening the stream. This only requires one system call, stat, and is
thus superior--at least in considering the number of system calls.
In particular, we want to prevent requesting a snap in the scenario that a wheel
block gets prevent-defaulted. In general the function is really only relevant
for touch input and there's no need to run it for other types of input.
--HG--
extra : commitid : oBr4VmShHv
When a make target is generated with FileAvoidWrite, this can cause targets to
get rebuilt perpetually when a prerequisite is updated, because FileAvoidWrite
will leave the target's mtime older than the prerequisite's when the target's
contents are unchanged. To avoid this issue, GENERATED_FILES is modified to
unconditionally update its target's mtime.
--HG--
extra : commitid : 4k5e5rKtPZ2
Bug 118468 landed an option for FileAvoidWrite to always write to an output
file, whether or not the contents would be changed. This was to address a
problem caused by not updating mtimes when building GENERATED_FILES, but
undoes the purpose of FileAvoidWrite and isn't really necessary.
This is addressed in a subsequent commit by unconditionally updating
mtimes when processing GENERATED_FILES.
--HG--
extra : commitid : AfOhgUstokq
GENERATED_FILES and accessible/xpcom/Makefile.in add to EXTRA_MDDEPEND_FILES, but for
targets that run during export. Export doesn't include EXTRA_MDDEPEND_FILES, so none
of them is ending up with correct dependencies. The EXTRA_EXPORT_MDDEPEND_FILES variable
could be used for this purpose, but given the circumstances this variable is removed,
and EXTRA_MDDEPEND_FILES is instead included unconditionally.
--HG--
extra : commitid : 7daRRnxfkS0
nsTextFrame didn't use the composition underline color if the
composition didn't have a foreground color defined. This patch makes it
use the underline color if foreground color is not defined.
GeckoEditable currently synthesizes key events when committing strings,
to improve web compatibility. However, synthesizing keys is causing
performance problems when the string is long. This patch limits key
synthesis to single-character strings. For longer strings, we commit the
string as a whole. We have other ways of ensuring web compatibility now
(sending dummy key events), so this restriction should not cause
regressions.
Currently, GeckoEditable periodically fires update composition events to
update the Gecko composition styling. To make the code more efficient
and more robust in dealing with content JS code, this patch merges these
events into events like replacing text, setting span, and removing span.
As a result, a setComposingText call now results in one replacing text
event instead of a replacing text event plus an update composition event.
Currently, the Gecko-side selection tries to mirror the Java-side
selection, but this is difficult because the Gecko selection has some
subtle but important behavior differences from the Java selection.
This patch separates out Gecko selection from Java selection so that
they're only loosely coupled. The two selections will periodically
synchronize through events and notifications, but at certain times they
may fall out-of-sync, for example when a composition is active. This
shouldn't affect functionality in a major way, and it's an acceptable
trade-off.