Move InspectorBackendDispatcher::commandNames array from .data to .rodata section

Move InspectorBackendDispatcher::commandNames array from .data to .rodata section

Move InspectorBackendDispatcher::commandNames array from .data to .rodata section
by getting rid of the pointers / relocations and using 1 single const char array
for all command names. An index array was introduced to store the index of each
command name in the char array.

This effectively moves commandNames array from .data to .rodata:
- Before: (.rodata: 8788 bytes / .data: 880 bytes)
- After: (.rodata: 9669 bytes / .data: 0 bytes)

The binary size is up 1 byte but we moved 880 bytes to shareable readonly memory,
thus potentially reducing RAM usage on Android a bit.

This change also reduces the number of relocations by ~220 according to relinfo,
which decreases the load time of the library.

BUG=249746
Committed: https://src.chromium.org/viewvc/blink?view=rev&revision=160734

[Inactive, 2013-10-28 17:01:33]

Diff on the generated InspectorBackendDispatcher.cpp:
http://pastebin.com/SswEN57a

Unfortunately, this makes the generator code a little bit more complex and it
needs to generate an additional array.

[pfeldman, 2013-10-28 17:12:36]

lgtm. But I am not sure it was worth the effort.

[Inactive, 2013-10-28 17:18:03]

On 2013/10/28 17:12:36, pfeldman wrote:
> lgtm. But I am not sure it was worth the effort.

One alternative would be to use const pointers. This would be a very minimal
change and would move the data to .data.rel.ro instead .rodata. The data could
still potentially be shared on Android thanks to the "crazy" linker. However,
the generated code would not be optimal as it would not get rid of the load-time
relocations.

Maybe it is an acceptable compromise though?

[digit1, 2013-10-28 17:27:34]

lgtm

On 2013/10/28 17:18:03, Chris Dumez wrote:
> One alternative would be to use const pointers. This would be a very minimal
> change and would move the data to .data.rel.ro instead .rodata. The data could
> still potentially be shared on Android thanks to the "crazy" linker. However,
> the generated code would not be optimal as it would not get rid of the
load-time
> relocations.
> 
> Maybe it is an acceptable compromise though?

Please do not create more constant pointers :-) They require relocations, that
bloat the binary, and make it slower to load. Even if the crazy linker can share
them, all the RELRO section pages have to be paged-in at load time. Putting
things in .rodata is much better because it will be loaded on demand. For rarely
used features, this is a net win.

Your change is actually the kind of things we want more in the Blink codebase!

[commit-bot: I haz the power, 2013-10-28 17:50:34]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/ch.dumez@samsung.com/48743004/60001

[commit-bot: I haz the power, 2013-10-28 18:39:25]

Change committed as 160734