Closed Bug 533043 Opened 12 years ago Closed 12 years ago

Use a single preprocessed packaging manifest for all three platforms

Categories

(Thunderbird :: Build Config, defect, P1)

defect

Tracking

(Not tracked)

RESOLVED FIXED
Thunderbird 3.1a1

People

(Reporter: philor, Assigned: philor)

References

(Blocks 1 open bug)

Details

Attachments

(2 files, 5 obsolete files)

Two primary benefits:

* because Fx uses a manifest on all three platforms, core hackers have no expectation that throwing some random piece of junk in dist/bin will result in it being shipped, while we have no system to ensure we look at and decide whether to unship new pieces of junk thrown in there

* because the patch to link XPTs on OS X without a manifest didn't land on 1.9.2, we would need yet another chunk of ifdef hell in our removed-files.in to deal with linked XPTs in 1.9.1+ on Windows, 1.9.2+ on Linux, and 1.9.3+ on Mac
Flags: blocking-thunderbird3.1+
Attached patch WIP v.1 (obsolete) — Splinter Review
Several more rounds to go, for 1.9.2 and shared and diffing against what we're packaging now, but this gives fairly reasonable package-compare results on 1.9.3 on all three platforms.
Attached patch Fix v.1 (obsolete) — Splinter Review
This seems to work, though package-compare is a bit ugly since for some reason I don't quite get, BINPATH starts with a / on OS X, and I didn't quite figure out how to compare both the stuff above Contents/MacOS/ and the stuff within, but it's still way better than the nothing we have now (as it points out that for some reason, despite the |MOZ_OJI=| in confvars.sh, we're still building OJI).

I've got a little Stockholm Syndrome from spending so much time with it, building six times for every change, so I'll attach the last six package-compares, too.
Attachment #416225 - Attachment is obsolete: true
Attachment #416627 - Flags: review?(bugzilla)
Attached file package-compares
I would appreciate if you could also mind Calendar, as soon as you have the issues sorted out for Tb :)
Attached patch Fix v.2 (obsolete) — Splinter Review
Dolske's been a busy little beaver - added components/crypto-SDR.js from bug 499417.
Attachment #416627 - Attachment is obsolete: true
Attachment #417432 - Flags: review?(bugzilla)
Attachment #416627 - Flags: review?(bugzilla)
Attached patch Fix v.3 (obsolete) — Splinter Review
Go go gadget hg pull --rebase, since of course crypto-SDR.js and mozsqlite would need to land and rot the source of my hg mv.
Attachment #417432 - Attachment is obsolete: true
Attachment #417870 - Flags: review?(bugzilla)
Attachment #417432 - Flags: review?(bugzilla)
Some thoughts on http://build.mozillamessaging.com/buildbot/try/builders/Try%20server%20comm-central%20mac%20hg%20builder/builds/280/steps/package%20compare/logs/stdio and http://build.mozillamessaging.com/buildbot/try/builders/Try%20server%20comm-central%20linux%20hg%20builder/builds/278/steps/package%20compare/logs/stdio and http://build.mozillamessaging.com/buildbot/try/builders/Try%20server%20comm-central%20win32%20hg%20builder/builds/277/steps/package%20compare/logs/stdio

* I forget exactly where, maybe the Fx single-pkg-manifest bug, but Ted fixed Packager.mk to properly deal with recursively copying a directory, so no more foo/* and no more modules/* modules/gloda/* (which is nice, since crashreporter.app/ would get ugly)

* somebody who's probably me should really deal with the way that we build run-mozilla.sh on OS X, where we have absolutely no expectation that it will work or desire to ship it

* I finally got the story on dependentlibs.list, which is a convenience for an XULRunner app which finds you as a GRE and wants to load you up, so once I get done wrestling with the makefile trying to not create it for non-libxul and add it to Fx's manifest, I'll add it to ours in a libxul ifdef to remind us why we're not shipping it

* After I don't know how many times thinking it was an annoying artifact of sort that put bin/icons/updater.png so far away from where I was packaging with *.png, I finally realized that duh, bin/chrome/icons/*.png isn't going to package anything in bin/icons, so I guess I'll be doing another patch once I figure out why Fx is packaging bin/icons/*.xpm when there aren't any
Attached patch Fix v.4 (obsolete) — Splinter Review
With updater.png, and without the redundant isp/@AB_CD@/.
Attachment #417870 - Attachment is obsolete: true
Attachment #418563 - Flags: review?(bugzilla)
Attachment #417870 - Flags: review?(bugzilla)
Attached patch Fix v.4.0001Splinter Review
Well, the part about isp/@AB_CD@/ would have been true, if I'd remembered to qref.
Attachment #418563 - Attachment is obsolete: true
Attachment #418603 - Flags: review?(bugzilla)
Attachment #418563 - Flags: review?(bugzilla)
Attachment #418603 - Flags: review?(bugzilla) → review+
http://hg.mozilla.org/comm-central/rev/c87ebb61ec41
Status: ASSIGNED → RESOLVED
Closed: 12 years ago
Resolution: --- → FIXED
Blocks: TB2SM
Flags: in-testsuite-
Flags: in-testsuite-
I've just checked in two packaging bustage fixes:

http://hg.mozilla.org/comm-central/rev/a5bd970f674d

Adds * onto the end of directories so that the contents gets packaged on Windows
Also adds thunderbird.xpt onto the removed-files.in list (I suspect this is mac only, and we might want to fix that in a follow-up, however it busted my trunk mac nightly build on update hence I wanted to add it).

http://hg.mozilla.org/comm-central/rev/8cc5f5213941

Adds the two modules/ sub-directories. Apparently recursive packaging doesn't work in windows or not in the way we think it does.

These are quick bustage fixes so that we don't break our builds too badly over Christmas, I'll let Phil work out if they need extra follow-ups or not.
Bleah, yeah: I'd forgotten that no-manifest linking uses the app name, since it doesn't have a manifest to get the component name from, so thunderbird.xpt is Mac and Linux. I actually did ask Ted why Fx's manifest uses * sometimes and not others, but neither one of us quite managed to realize that it uses * when it's not a Mac-only thing, and only uses just / in Mac ifdefs.
Depends on: 537593
You need to log in before you can comment on or make changes to this bug.