785175 - Nightly can no longer run BananaBread

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[(no longer active)]

STR: Load http://dl.dropbox.com/u/80664946/main.html?medium,medium  The game gets stuck during the Preparing stage.

I see this on the error console:
[14:51:10.372] dc[y[(Gc >> 2)]] is not a function @ http://dl.dropbox.com/u/80664946/crunch-worker.js:110

Followed by a whole lot of errors such as:
[14:51:10.372] uncaught exception: Unsupported image format undefined for packages/gk/fantasy/stone_ground_gk_v01/stone_ground_gk_v01_cc.dds

Comment 1

[(no longer active)]

Note that this regression doesn't exist on my 8/20 nightly.

Comment 2

[(no longer active)]

Regression range: http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=360ab7771e27&tochange=abc17059522b

Comment 3

[Alice0775 White]

Regression window(m-c)
Good:
http://hg.mozilla.org/mozilla-central/rev/e33c175f6807
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:17.0) Gecko/17.0 Firefox/17.0 ID:20120821162246
Bad:
http://hg.mozilla.org/mozilla-central/rev/abc17059522b
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:17.0) Gecko/17.0 Firefox/17.0 ID:20120822030558
Pushlog:
http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=e33c175f6807&tochange=abc17059522b


Regression window(m-i)
Good:
http://hg.mozilla.org/integration/mozilla-inbound/rev/0fa2e09c092e
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:17.0) Gecko/17.0 Firefox/17.0 ID:20120821080500
Bad:
http://hg.mozilla.org/integration/mozilla-inbound/rev/57c1c330e85f
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:17.0) Gecko/17.0 Firefox/17.0 ID:20120821090701
Pushlog:
http://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=0fa2e09c092e&tochange=57c1c330e85f

Comment 4

[Luke Wagner [:luke]]

This is looking like a dup, I'll verify shortly.

Comment 5

[Luke Wagner [:luke]]

Ok, the error during "Preparing" looks like it is fixed by the patch in bug 784639 which isn't yet on nightlies.

Comment 6

[Luke Wagner [:luke]]

(which was indeed caused by the bug fingered in comment 3)

Comment 7

[Luke Wagner [:luke]]

Coincidentally, a trunk (opt and debug) build produces a new error (very early, before downloading is finished) about a switch statement being too big.  Seems unrelated, but I'll look into it.

Comment 8

[Luke Wagner [:luke]]

So the "switch too large" bug is from the just-landed bug 568142.  To be clear, the bug currently on nightly is fixed by bug 784639 and the current problem I'm seeing on trunk is not yet on nightly.

Alex: we seem to have a file with a very very long column (colspan = 4492669 at the point of the error); can we have a better fallback than reporting an error?  (This seems like a general problem for large minified JS programs.)

Comment 9

[u443197]

Attachment: Discard column information if the script is too large instead of aborting

Comment 10

[Alon Zakai (:azakai)]

Just wanted to say thanks to everyone here for the quick and effective responses. We are about to launch BananaBread in a few days and two breakages on Nightly today was very worrying!

I hope we can get the patch in comment 9 landed ASAP.

Comment 11

[(no longer active)]

Hmm, can we back out bug 568142 and reland it with this fix?  I've been meaning to do a blog post about profiling JS heavy applications and I have been meaning to use the experimental bananabread builds as an example, but this is blocking me, and my goal was to have that posted this week... :/

Comment 12

[Alon Zakai (:azakai)]

+1 to comment 11, if we can't land that fix immediately I support backing the breaking bug 568142 out. We are about to launch BananaBread in a few days and this can block us if it isn't fixed.

Comment 13

[Jason Orendorff [:jorendorff]]

Comment on attachment 654811 [details] [diff] [review]
Discard column information if the script is too large instead of aborting

Review of attachment 654811 [details] [diff] [review]:
-----------------------------------------------------------------

::: js/src/jit-test/tests/basic/bug785175.js
@@ +5,5 @@
> +}
> +str += 'var a = 1 + 1;';
> +
> +// don't throw an exception even though the column numbers cannot be maintained
> +eval(str);

Could also test that the line number is correct even though the column info is discarded. r=me with or without that test.

Comment 14

[u443197]

https://hg.mozilla.org/integration/mozilla-inbound/rev/e5259919146b

Comment 15

[(no longer active)]

I landed this on central directly so that I can trigger a nightly on it.  Thanks for your patch, Alex!

http://hg.mozilla.org/mozilla-central/rev/f2146a6c104e

Comment 16

[Jim Blandy :jimb]

Stuff like this is what makes this job exciting. "A source file with a four million column line?" Ran into one of those within a week!"

Comment 17

[Alon Zakai (:azakai)]

Thanks Alex, Jason and Ehsan for the quick work here! Very happy the next Nightly will be ok on this issue.

Comment 18

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/e5259919146b