28841 - space-minimize nsStr structure

Status: VERIFIED FIXED

(Core :: XPCOM, defect, P3)

Description

[Chris Waterson]

Several alternatives:

1. Just pack bits and use less space. (jevering changed lengths to 16 bits, 
etc., and was able to reduce the app's core size by 200Kb, or 1%).

2. Use a more extravagant packing scheme, which would make nsStr just point to a 
buffer, the buffer which would contain the fields as a prefix.

This work should be accompanied by demonstratable space results.

Comment 1

[Warren Harris]

Somebody explain to me why we need the mCharsize field at all. The 2 subclasses 
of nsStr, nsString and nsCString, know the charsize implicitly, don't they?

Comment 2

[Alec Flett]

Once this is done, I'll do the work to roll nsXPIDL*String into ns*String
Filing a new bug now

Comment 3

[Chris Waterson]

alecf: there is already a bug open for that work. see bug 16108.

Comment 4

[Suresh Duddi (gone)]

Scott ?

Comment 5

[Scott Collins]

Chris is instrumenting |nsString| to help determine which alternative will yield 
the greatest gains.  See

  <http://bugzilla.mozilla.org/show_bug.cgi?id=28842>

A good measure might be to add (for profiling only, of course) a time field to 
the string structure, and log how _long_ a string exists at a certain size.  That 
way we could tell the difference between having lots of empty strings and having 
lots of strings that just happen to start empty but don't really spend much time 
in that state.

Alec, with respect to |getter_Copies|, |getter_CopiesWithCharset|, et al, I 
appreciate the help :-)  The notion of combining |nsXPIDL[C]String|s 
functionality directly into |nsString|, and eliminating the other classes is a 
major value pushing for the second alternative above (since it gives us more 
flexibility with the length, and definitely will find the places where people are 
using it in ways that interfere with this plan).

Comment 6

[Scott Collins]

cc'ing jevering at his request.  jevering also reports that rickg is examining 
these issues as well.

Comment 7

[rickg]

I'm taking this from you scott; I agree with the analysis, and will save you the 
trouble, so that you can focus on bigger issues.

Comment 8

[Scott Collins]

Yeah, when this bug was first filed I talked to jevering about whether I should 
re-assign it to you.  His argument was similar, but in the opposite direction.  
He wanted me to keep the bug, since you necessarily are focusing on the bigger 
issues of management, directorship, and the PDT team.  I'll keep the bug, thanks.

Comment 9

[Scott Collins]

It looks like you are doing this work, rick, as part of the |StringValue| stuff. 
So you should own the bug in spite of earlier comments.  Sorry, and thank you.

Comment 10

[rickg]

Ok -- so nsStr is going away altogether, and the new string is alrady compressed 
(to what I think is the min size). I'm closing this as a result of that work, 
already in progress, as it refers to a design goal that (I think) I'm meeting in 
the new approach.

Comment 11

[leger]

Marking Verified for his bug per rickg comments.