There are, sadly, many combinations of linkage in use throughout the tree.
The main differentiator, though, is between program/libraries related to
Gecko or not. Kind of. Some need mozglue, some don't. Some need dependent
linkage, some standalone.
Anyways, these new templates remove the need to manually define the
right dependencies against xpcomglue, nspr, mozalloc and mozglue
in most cases.
Places that build programs and were resetting MOZ_GLUE_PROGRAM_LDFLAGS
or that build libraries and were resetting MOZ_GLUE_LDFLAGS can now
just not use those Gecko-specific templates.
This hack has actually not been actively used since sqlite, nss and nspr are
all folded together, because no shared library is actually linked in
db/sqlite3/src.
Since essentially everything is linked to libmozglue and libmozglue takes
precedence in symbol resolution in our dynamic linker, there is no need
to wrap most symbols. PR_GetEnv/PR_SetEnv still needs wrapping because
there's no other way to actually wrap the calls from NSPR itself and NSS,
as well as the symbols wrapped because our dynamic linker can't find them
in system libraries on some devices because they're weak.
Ever since bug 969164, the js build system, when building gecko (not when
building standalone) uses a autoconf-js.mk file for its config.
One of the suboptimal ways we have to retrigger builds when the build
configuration changes (changes to e.g. configure.in can do that) is to
make most things depend on autoconf.mk. Which unfortunately doesn't
account for the fact the js/src subdirectory uses a different file.
In practice, this means that some classes of changes to the js build
system, not accompanied with toplevel build system changes may no
trigger the corresponding rebuilds in the js subtree on incremental
builds.
Currently, when there is both an expandlibs descriptor and an actual static
library, expandlibs picks the static library. This has the side effect that
if there are object files in the static library that aren't directly used,
they're dropped when linking, even when they export symbols that would be
exported in the final linked binary.
In most cases in the code base, files are not dropped that way. The most
notable counter-example is xpcomglue, where actually not dropping files
leads to link failure because of missing symbols those files reference
(yes, that would tend to say the glue is broken in some way).
On the opposite side, there is mozglue, which does have both a descriptor
and a static library (the latter being necessary for the SDK), and that
linking as a static library drops files that shouldn't be dropped (like
jemalloc). We're currently relying on -Wl,--whole-archive for those files
not to be dropped, but that won't really be possible without much hassle
in a world where mozglue dependencies live in moz.build land.
Switching expandlibs to use descriptors when they exist, even when there
is a static library (so, the opposite of the current behavior) allows to
drop -Wl,--whole-archive and prepare for a better future. However, as
mentioned, xpcomglue does still require to be linked through the static
library, so we need to make it a static library only.
To achieve that, we make NO_EXPAND_LIBS now actually mean no expandlibs
and use that to build the various different xpcomglues.
Incidentally, this makes the build fail because stdc++compat.o then appears
twice on the libxul linkage command line. This, in turn, is because
widget/xremoteclient creates both a program and an intermediate library for
libxul. The Program() template adds stdc++compat to the directory data,
which ends up being added to libxul as well. The same kind of issue arises
when linking the gtest-enabled libxul.
While eventually we'll be able to avoid those problems, it's not the case
yet, so work around the issue by making expand-libs skip .desc files that
appear multiple times.
