474073 - Make first run non-xhtml

Status: VERIFIED WONTFIX

(Firefox for Android Graveyard :: General, defect)

Description

[(dormant account)]

Taking out the dtd saves 200ms of startup.

Comment 1

[(dormant account)]

Attachment: avoid the dtd

Switch to using a js file instead of dtd and to plain html instead of xhtml.

Comment 2

[(dormant account)]

Attachment: avoid dtd (with firstrun.js)

missed firstrun.js in the previous patch

Comment 3

[Mark Finkle (:mfinkle) (use needinfo?)]

Comment on attachment 358036 [details] [diff] [review]
avoid dtd (with firstrun.js)

I just bitrot this patch pretty good. Rework itand we can get it landed.

Comment 4

[:Gavin Sharp [email: [e-mail]]]

We'll have to either figure out a different way to do this, or do some localizer outreach to make sure they understand how to localize the file, but I think this approach can work for now.

Comment 5

[Axel Hecht [:Pike]]

We don't put js files into l10n. That has both security reasons, and l10n policy reasons.

Security as in that we don't trust our localizers to contribute arbitrary js code to our product, and l10n policy in that the number of file formats we expose to localizers is limited.

I don't think that a defines file would work either, that would only fix the policy part, but not the code insertion piece.

We've seen l10n taking other startup hits before, i.e., just having en-US.jar etc seems to have an impact on TP, but I don't see us ripping that out.

IMHO, this should be WONTFIX.

Comment 6

[(dormant account)]

(In reply to comment #5)
> We don't put js files into l10n. That has both security reasons, and l10n
> policy reasons.

Appreciate detailed feedback. I guess our alternatives are to
a) Fix nsFileChannel-related code so it doesnt force a mimetype check on .dtd files
b) stick xhtml11.dtd into a jar

Comment 7

[Zibi Braniecki [:zbraniecki][:gandalf]]

did we consider hit difference from .properties? (I expect it to be higher than .dtd, but maybe I'm wrong?)

Also, opeion a) sounds pretty reasonable.

Comment 8

[Axel Hecht [:Pike]]

Gandalf, the perf hit seems to only half-way come from the l10n dtd, but also from the xhtml11.dtd we load from the res:// protocol. Which we do for all xhtml files. Which makes a) a good thing to fix no matter which.

Comment 9

[:Gavin Sharp [email: [e-mail]]]

The JS code here isn't being included in the product, it's only being run in the unprivileged firstun page, for what it's worth.

Comment 10

[Joel Maher ( :jmaher ) (UTC -8)]

marking as verified