33782 - Fix lower/upper case and alias resolution issues for charset names

Status: RESOLVED WONTFIX

(Core :: Internationalization, defect, P3)

Description

[cata]

Lately we were hit by a few bugs generated by the lowercase/uppercase 
representation of the charset names. It's time to fix this.

The rule is to use only cannonical charset names everywhere. This will be 
achieved by using a special object to identify a charset for all purposes. This 
object can only be obtained by calling a certain API. This API will do alias 
resolution. The objects will then be unique, like Atoms, so a comparison will 
only mean comparing pointers.

Steps:
1) Create the interface for the charset object.
2) Create the API creating these objects.
3) Move people to use this new API.
4) Deprecate old API(s).

Possible issue I need to look into: ProgID's are also case sensitive.
Future impovement thought: nicer alias properties file format.

Comment 1

[bobj]

Added bugs that depend upon this.
When will this land?  If not real soon, let's provide some workarounds for
the crashing bugs that depend upon this.

Comment 2

[cata]

This will land real soon. (1-2 days) But the work is being done on the tip, so 
it won't fix the beta1 crashes...

Comment 3

[bobj]

Tip is fine.
1-2 days is to complete the API or to migrate all code to use the new API?
Do you know how many places need to be changed?

Comment 4

[cata]

In an attempt to minimise the amount of necessary work and time, I'll try to use 
nsIAtom as the charset type. That means that the interface & the new API are 
already there. What's left to do is just a fix on the API implementation (ready 
in my tree) and then step 3&4.

Possible issues when using nsIAtom instead of an "nsICharset":
- maybe some performance loss from going to the Atom engine
- people creating directly the Atom instead of going through GetCharsetAtom() 
If these problems will hurt us, we can switch to nsICharacterSet at any time.

Now, the idea for a so-called Step 6) for the future: improvement of the 
GetCharsetAtom() implementation. Right now, it is build on top of two things: 
the nsCharsetAlias service and the Atom engine. We can improve on both. First, 
make the nsCharsetAlias better, with its own hash table (not the StringBundle 
anymore) and a nicer file format. Second, reimplement the Atom engine for the 
specific Charset case - more optimal. Finally, unify the two for even better 
performance.

Comment 5

[Frank Tang]

Mark it as won't fix.