Everything is in the summary
Created attachment 521078 [details] [diff] [review] Patch python code is current trunk on your pyxpt repo. As for the config.mk/rules.mk changes, I didn't think there was much interest in complicating things to keep the dependency for xpt files on XPIDL_LINK, the error message in case XPIDL_LINK doesn't work should be as explicit as a dependency miss.
Comment on attachment 521078 [details] [diff] [review] Patch The build bits look fine, but since I wrote all that Python code, someone else needs to give this a once-over. I don't know if bsmedberg has the time or inclination to review it, but if not, perhaps catlee would?
Created attachment 521759 [details] [diff] [review] Test patch I used this patch to try to see how things differ, and I hit a breakage pretty early: cmp _xpidlgen/xpcom_base.xpt _xpidlgen/xpcom_base-new.xpt _xpidlgen/xpcom_base.xpt _xpidlgen/xpcom_base-new.xpt differ: char 24, line 3 make: *** [_xpidlgen/xpcom_base.xpt] Error 1 However, if I xpt_dump both, the output is exactly the same.
If I don't compare the files directly but compare dumps instead, I can go further, but stop again later: --- obj-i686-pc-linux-gnu/xpcom/components/_xpidlgen/xpcom_components.xpt.dump 2011-03-25 09:34:00.702062243 +0000 +++ obj-i686-pc-linux-gnu/xpcom/components/_xpidlgen/xpcom_components-new.xpt.dump 2011-03-25 09:34:00.702062243 +0000 @@ -8,8 +8,6 @@ Interface Directory: - ::nsIFile (00000000-0000-0000-0000-000000000000): [Unresolved] - - ::nsISimpleEnumerator (00000000-0000-0000-0000-000000000000): - [Unresolved] - ::nsISupports (00000000-0000-0000-c000-000000000046): [Unresolved] - ::nsIFactory (00000001-0000-0000-c000-000000000046): @@ -114,3 +112,5 @@ uint32 createInstanceByContractID(in string, in nsISupports, in nsIID &, out retval InterfaceIs *); Constants: No Constants + - ::nsISimpleEnumerator (d1899240-f9d2-11d2-bdd6-000064657374): + [Unresolved]
I investigated this, and it looks harmless. My patch from bug 455512 ( http://hg.mozilla.org/mozilla-central/rev/a78f9bb9006e ) attempted to make xpt_link prefer unresolved interface descriptors with a valid IID, but apparently that's broken and it prefers the ones with zero IIDs. xpt.py gets it right, so that's the only difference, This difference disappears in the final linked xpt, since all interfaces are resolved at that point. I think we simply declare the xpt_link binary buggy and move on in life.
Ftr, http://hg.mozilla.org/comm-central/rev/e11c23c9a55b "Fix xpt_link bustage from bug 643817 as it got replaced with xpt.py."