Now that we're moving to short-cycle releases without binary compatibility guarantees, the system-XULRunner support in the standalone glue is not going to be useful, and will be actively dangerous. I think I'd also like to remove the standalone glue entirely, but glandium is apparently trying to use it to improve startup time somehow, so we might need to discuss that part further.
With my Debian developer hat on, the /etc/gre.d, ~/.gre.d thingy has always been pointless, even when releases cycles were long. So I'm all okay with that going away. The standalone glue is however useful to avoid using rpath on applications that do use xulrunner/libxul, such as kazehakase, galeon, etc. And this system registration is part of how it works. I'm all for changing the API, though (such as GRE_GetGREPathWithProperties). With my Mozilla hat on, I think the standalone glue is useful for us to avoid creating our own dynamic loader on each platform to overcome platform limitations and improve cold startup time without relying on a wrapper.
Created attachment 520237 [details] [diff] [review] Remove GRE finding and registration code, rev. 1
This covers everything except gtkmozembed, which I'll eal with separately.
Comment on attachment 520237 [details] [diff] [review] Remove GRE finding and registration code, rev. 1 Looks good to me. I guess applications using GRE_GetGREPathWithProperties today could use a semi-hardcoded path, taken from e.g. one of the variables provided by the pkg-config files on Unix.
Yes, Linux apps should know (somehow) where their target xulrunner is going to be located by the package manager or whatever.
This landed as http://hg.mozilla.org/mozilla-central/rev/d7fc46501527 but it caused linux bustage*, so I backed it out as http://hg.mozilla.org/mozilla-central/rev/84034146670c *bustage logs: http://tinderbox.mozilla.org/showlog.cgi?log=Firefox/1301679138.1301680151.17956.gz http://tinderbox.mozilla.org/showlog.cgi?log=Firefox/1301679130.1301680106.17885.gz http://tinderbox.mozilla.org/showlog.cgi?log=Firefox/1301679126.1301680748.20017.gz Bustage looks like e.g.: > TestGtkEmbedNotebook.cpp: In function 'int main(int, char**)': > TestGtkEmbedNotebook.cpp:17: error: 'GREVersionRange' does not name a type
This landed again as http://hg.mozilla.org/mozilla-central/rev/8ad43371c469
Created attachment 530075 [details] [diff] [review] remove the help too
The additional patch landed: http://hg.mozilla.org/mozilla-central/rev/54bacc404a6f
Backed out because something that landed with this may have caused a new orange (Win debug Moth tabview/browser_tabview_bug597248.js): http://hg.mozilla.org/mozilla-central/rev/eaaa24ff93a0 This can probably land again once we figure out the cause of the test failures.
Re-landed the followup patch: http://hg.mozilla.org/mozilla-central/rev/59ae97ec2f12
Does anyone know of any actual documentation for this feature that has been removed? I can't find any to update.
https://developer.mozilla.org/en/GRE and https://developer.mozilla.org/en/XPCOM_Glue both mention the removed function.
Yeah, I found those moments after posting my question. I take it this means that you need to know where to get your xpcomglue lib instead of having it hunted down for you?
Documentation updated (I found my answers by, you know, reading): https://developer.mozilla.org/en/GRE#Find_a_compatible_GRE https://developer.mozilla.org/en/XPCOM_Glue#Frozen_linkage.3a_standalone_glue_%28no_DLL_dependencies%29