It seems that if we're going to enable JS_SHARED_LIBRARY, we need ICU to be a
shared library too in order to prevent code duplication in mozjs.dll bloating
it to a whopping 21 MB.
This unfortunately means we have to fold icudata back to icu itself. But since
icudata is still separated from libxul (which is what really matters for #1451),
I think this is a fair compromise.
Building ICU as shared is offered as an option via --enable-shared-icu. But
technically you really need --enable-shared-icu if you're going to use
--enable-shared-js. Unfortunately I can't make configure exit with an error if
MOZ_SHARED_ICU is not defined when JS_SHARED_LIBRARY is enabled, because options
are controlled by different configure scripts (icu is based on bash, while js is
based on python), and I have no idea how to pass MOZ_SHARED_ICU to js's
moz.configure.
Tag #62
Don't mark the .kPStaticModules section as relro for bfd, as the linkers on
Windows and Mac don't do this anyway.
Add a section insert of .kPStaticModules to config/expandlibs_exec.py, which
seems to be the primary cause of linking bustage on gold, lld, and mold.
Based on Mozilla bugs 1270140, 1504656, 1483545, and 1504334.
Took note of Mozilla bug 1379222 which changed GetPrototypeFromCallableConstructor
to GetPrototypeFromBuiltinConstructor.
There are many other changes I did myself since the initial implementation by
Mozilla wouldn't work with this codebase.
This allows us to use a greater rendering depth for exceedingly-deep DOM trees
in layout, better matching what mainstream browsers are capable of. Note that
for 32-bit Windows the stack size MUST be set to larger than the default or Bad
Things Will Happen™ - we use 1.5 MB for this as a carefully-tuned value.
This needs to be capped specifically for JS use because some JavaScript
obfuscators deliberately trigger stack overflows and would lock up the browser
otherwise as long as there's still stack space to abuse. For web compatibility
we therefore limit this to 2MB in JS only (3x for ASAN) while still allowing
a greater depth for the layout engine.
Including <cstdlib> must never trigger inclusion of "mozalloc.h", because the
first thing this one does is precisely to include <cstdlib>.
It is now the case that, in libc++ 9 and beyond, <cstdlib> includes <math.h>
which then tries to include <type_traits>, which the STL wrapper intercepts,
trying to load "mozalloc.h". Same problem happens with <limits>.
Fix pulled from: https://bugzilla.mozilla.org/show_bug.cgi?id=1594027.
This removes speech recognition, pocketsphinx, training models
and the speech automated test interface.
This also re-establishes proper use of MOZ_WEBSPEECH to work
for the speech API (synthesis part only) that was a broken mess
before, with some synth parts being always built, some parts
being built only with it enabled and recognition parts being
dependent on it. I'm pretty sure it'd be totally busted if you'd
ever have tried building without MOZ_WEBPEECH before.
Tested that synthesis still works as-intended.
This resolves#1538
This updates our behavior for computed DOM styling to no longer return
null on elements that have no display, but return a 0-length (empty)
style instead and don't throw. For this we stop looking at having a
presentation for the style and just look at the document instead.
This resolves#1219
