592323 - Fastload not invalidated after `make`ing on OS X

Status: RESOLVED FIXED

(Core :: XPCOM, defect)

Description

[Paul O'Shannessy [:zpao] (not bugmail, email directly)]

I chatted with Ben and he said running make should create a .purgecaches in obj/dist/bin, which it does. However on OS X we run MinefieldDebug.app and my fastload isn't invalidated, so I have to run it twice for my fastload to be invalidated.

Comment 1

[Benedict Hsieh [:benedict]]

Huh, didn't catch that earlier. So the situation is: make, run once (not invalidated), run again (now, invalidated)?

I thought the situation was that running from MinefieldDebug.app means we don't find the .purgecaches file, but seems maybe that's not the case. Anyways, I have an OSX setup so I can look into this myself also.

Comment 2

[Paul O'Shannessy [:zpao] (not bugmail, email directly)]

(In reply to comment #1)
> Huh, didn't catch that earlier. So the situation is: make, run once (not
> invalidated), run again (now, invalidated)?

Correct.

Comment 3

[Paul O'Shannessy [:zpao] (not bugmail, email directly)]

If you want to test my case as closely as possible...

add a line in browser/components/sessionstore/src/nsSessionStore.js in init...
dump("\n\nCAN'T MISS THIS LINE\n\n");

% make -C browser/components/sessionstore && make -C browser/app
% ./dist/MinefieldDebug.app/Contents/MacOS/firefox -P whatever

The dump won't show up the first time, but will the 2nd time you run.

.purgecaches does exist in dist/MinefieldDebug.app/Contents/MacOS (everything in dist/bin gets rsynced when running make in browser/app). It gets put in dist/bin after the first make so get's rsynced.

So perhaps it's looking for the file in the wrong place? It looks like it should be ok, unless there's something special that needs to be done on Mac...

Comment 4

[Benedict Hsieh [:benedict]]

Seems that the problem here is that
>nsDirectoryService::gService->Get(NS_XPCOM_CURRENT_PROCESS_DIR, ...)
is equivalent to
>gAppData->directory
on Linux but not on OSX. 

On OSX, the former returns [obj-dir]/dist/MinefieldDebug.app/Contents/MacOS, and the latter returns [obj-dir]/dist/MinefieldDebug.app/Contents/ Haven't tested on Windows, yet. Not sure if this is a bug in gAppData or intended.

nsDirectoryService isn't available when I'm checking for the .purgecaches file, so maybe I need to check for it later in nsXPComInit.cpp. I'll ask around, just sticking these notes in the bug for now.

Comment 5

[Benedict Hsieh [:benedict]]

Attachment: fix

Yikes, the above comment is wrong in every way. Some bad test code / patch queues led me astray.

I didn't investigate why it used to work on the second run, but this patch makes it work the first time around. I think that reason that this bug surfaced now instead of months ago, when I checked in the offending code, was that the old fastload dependency code was saving us.

Comment 6

[Benjamin Smedberg]

Comment on attachment 471582 [details] [diff] [review]
fix

Heh!

Comment 7

[Dão Gottwald [:dao]]

http://hg.mozilla.org/mozilla-central/rev/499c24ccb868