Build shared libraries by default with -shared

[sbc100]

The old behavior was to build fake/dummy shared libraries which were really just normal object files. This old behavior can still achieved by settings the new FAKE_DYLIBS settings.

[sbc100]

@dschuff @kripken not quite ready for review yet but what do you think of this general direction?

[kripken]

My vague concern is that in practice I recall various build systems used -shared unnecessarily. Ignoring it meant porting was easier (no need to figure out how to remove that link flag from the build system). That might have been just unlucky choices in the things I ported, though - do you have a sense how common the flag is?

[sbc100]

My vague concern is that in practice I recall various build systems used -shared unnecessarily. Ignoring it meant porting was easier (no need to figure out how to remove that link flag from the build system). That might have been just unlucky choices in the things I ported, though - do you have a sense how common the flag is?

True, yes, those are exactly the users who would want to start using -sFAKE_DYLIBS.

We could land this with -sFAKE_DYLIBS enabled by default (i.e. we could this essentialy a no-op) for existing users, and then we could propose changing the default as a followup? That would be safer I guess.

[dschuff]

Do we currently emit any kind of warning when a user specifies -shared but we ignore it?

[sbc100]

Do we currently emit any kind of warning when a user specifies -shared but we ignore it?

yes we have had that for a very long time:

$ emcc hello.c -shared -o libfoo.so
emcc: warning: linking a library with `-shared` will emit a static object file.  This is a form of emulation to support existing build systems.  If you want to build a runtime shared library use the SIDE_MODULE setting. [-Wemcc]

[dschuff]

I guess that's good because it raises awareness of the issue, but maybe bad if those folks then just go and add -Wno-emcc and then if we start a deprecation/upcoming behavior change warning they might miss it :)

[sbc100]

I guess that's good because it raises awareness of the issue, but maybe bad if those folks then just go and add -Wno-emcc and then if we start a deprecation/upcoming behavior change warning they might miss it :)

Or folks that run cmake the don't read all the warnings.

[sbc100]

OK, I'm going to split of the FAKE_DYLIBS setting into a new change that can be landed without changing any existing behaviour.