96529 - delay loading of unicharutil in nsString case conversion code

Status: RESOLVED FIXED

(Core :: XPCOM, defect)

Description

[Frank Tang]

this is a spin off from 75041
I find out one of the reason unicharutil get load into memory is nsString case
conversion. I have a fix for it and will include later for scc to review.

Comment 1

[Frank Tang]

Attachment: patch to reduce the need of loading unicharutil in init time to speed up startup perf

Comment 2

[Frank Tang]

with this one and 96530, we can remove the loading of unicharutil.dll with
-chrome blank.xul. 

Comment 3

[Frank Tang]

add perf

Comment 4

[Frank Tang]

add jbetak to the cc list. 

Comment 5

[Chris Waterson]

Attachment: lazily instantiate gCaseConv without scanning string first

Comment 6

[Chris Waterson]

How about something like attachment 47922 [details] [diff] [review]? This basically does the same thing as
your patch, ftang, but avoids scanning the string first. It proceeds using the
ASCII to upper/lower conversion until it hits a non-ASCII character. At that
point, it faults in the unicode utilities and lets it process the remainder of
the string. (Check my math to make sure I got the length parameter right.)

Comment 7

[Frank Tang]

while (cp <= end) {
....
+      result = PRInt32(aToUpper
+                       ? gCaseConv->ToUpper(cp, cp, end - cp)
+                       : gCaseConv->ToLower(cp, cp, end - cp));

while cp == end, it will go into the loop and end-cp will be equal to 0. 
Is that right? should it be end - cp + 1 ?

Comment 8

[Frank Tang]

*** Bug 98222 has been marked as a duplicate of this bug. ***

Comment 9

[Chris Waterson]

ftang: you are right! But perhaps a cleaner fix might be to simply let |end|
point one past the end of the string, and change the loop termination condition
to ``cp strictly less-than end''. Let me attach a patch...

Comment 10

[Chris Waterson]

Attachment: make |end| point one past end of the string, alter loop condition to strictly less-than

Comment 11

[Frank Tang]

Comment on attachment 48230 [details] [diff] [review]
make |end| point one past end of the string, alter loop condition to strictly less-than

r=ftang
It looks better.

Comment 12

[Frank Tang]

seeking for sr

Comment 13

[Brendan Eich [:brendan]]

Comment on attachment 48230 [details] [diff] [review]
make |end| point one past end of the string, alter loop condition to strictly less-than

/me grumbles about selective respect for the 80 column limit.

sr=[e-mail]

Comment 14

[Frank Tang]

fixed and check in