Add support for startup data (snapshot) compression.

Add support for startup data (snapshot) compression.

This is for mobile platforms where application footprint size is
important. To avoid including compression libraries into V8, we assume
that the host machine have them (true for Linux), and rely on embedder
to provide decompressed data.

Currently, only snapshot data can be comressed. It is also possible to
compress libraries sources, but it is more involved and will be
addressed in another CL.

BUG=none
TEST=none

Committed: http://code.google.com/p/v8/source/detail?r=7724

[Søren Thygesen Gjesse, 2011-04-29 06:50:29]

LGTM

http://codereview.chromium.org/6901090/diff/1/SConstruct
File SConstruct (right):

http://codereview.chromium.org/6901090/diff/1/SConstruct#newcode343
SConstruct:343: },
As this is Linux only I think this should be either inside the os:linux section
above or have a 'os:linux' inside. I think an os:linux here is the best choice,
and makes adding more platforms most logical.

http://codereview.chromium.org/6901090/diff/1/SConstruct#newcode488
SConstruct:488: 'CPPDEFINES':   ['COMPRESS_STARTUP_DATA_BZ2'],
Ditto for the LIBS part.

http://codereview.chromium.org/6901090/diff/1/include/v8.h
File include/v8.h (right):

http://codereview.chromium.org/6901090/diff/1/include/v8.h#newcode2646
include/v8.h:2646: int id;
Is this id actually required?

http://codereview.chromium.org/6901090/diff/1/include/v8.h#newcode2649
include/v8.h:2649: int decompressed_size;
Should the compression algorithm be exposed here (enum or string)?

http://codereview.chromium.org/6901090/diff/1/include/v8.h#newcode2682
include/v8.h:2682: * The following 3 functions are to be used when V8 is built
with
with -> with the

http://codereview.chromium.org/6901090/diff/1/samples/shell.cc
File samples/shell.cc (right):

http://codereview.chromium.org/6901090/diff/1/samples/shell.cc#newcode31
samples/shell.cc:31: #ifdef COMPRESS_STARTUP_DATA_BZ2
Should this support be added to D8 as well? The process sample? For the process
sample maybe just add an #error.

http://codereview.chromium.org/6901090/diff/1/src/mksnapshot.cc
File src/mksnapshot.cc (right):

http://codereview.chromium.org/6901090/diff/1/src/mksnapshot.cc#newcode130
src/mksnapshot.cc:130: bool Compress(Compressor* compressor) {
ASSERT_EQ(-1, decompressed_size_)?

http://codereview.chromium.org/6901090/diff/1/src/snapshot-empty.cc
File src/snapshot-empty.cc (right):

http://codereview.chromium.org/6901090/diff/1/src/snapshot-empty.cc#newcode36
src/snapshot-empty.cc:36: 
How about removing decompressed_ and instead adding raw_ (or something like
that) to the input data, e.g.

const byte Snapshot::raw_data_[] = { 0 };
const byte* Snapshot::data_ = NULL;

[Søren Thygesen Gjesse, 2011-04-29 06:55:58]

On 2011/04/29 06:50:29, Søren Gjesse wrote:
> LGTM
> 
> http://codereview.chromium.org/6901090/diff/1/SConstruct
> File SConstruct (right):
> 
> http://codereview.chromium.org/6901090/diff/1/SConstruct#newcode343
> SConstruct:343: },
> As this is Linux only I think this should be either inside the os:linux
section
> above or have a 'os:linux' inside. I think an os:linux here is the best
choice,
> and makes adding more platforms most logical.
> 
> http://codereview.chromium.org/6901090/diff/1/SConstruct#newcode488
> SConstruct:488: 'CPPDEFINES':   ['COMPRESS_STARTUP_DATA_BZ2'],
> Ditto for the LIBS part.
> 
> http://codereview.chromium.org/6901090/diff/1/include/v8.h
> File include/v8.h (right):
> 
> http://codereview.chromium.org/6901090/diff/1/include/v8.h#newcode2646
> include/v8.h:2646: int id;
> Is this id actually required?
> 
> http://codereview.chromium.org/6901090/diff/1/include/v8.h#newcode2649
> include/v8.h:2649: int decompressed_size;
> Should the compression algorithm be exposed here (enum or string)?
> 
> http://codereview.chromium.org/6901090/diff/1/include/v8.h#newcode2682
> include/v8.h:2682: * The following 3 functions are to be used when V8 is built
> with
> with -> with the
> 
> http://codereview.chromium.org/6901090/diff/1/samples/shell.cc
> File samples/shell.cc (right):
> 
> http://codereview.chromium.org/6901090/diff/1/samples/shell.cc#newcode31
> samples/shell.cc:31: #ifdef COMPRESS_STARTUP_DATA_BZ2
> Should this support be added to D8 as well? The process sample? For the
process
> sample maybe just add an #error.
> 
> http://codereview.chromium.org/6901090/diff/1/src/mksnapshot.cc
> File src/mksnapshot.cc (right):
> 
> http://codereview.chromium.org/6901090/diff/1/src/mksnapshot.cc#newcode130
> src/mksnapshot.cc:130: bool Compress(Compressor* compressor) {
> ASSERT_EQ(-1, decompressed_size_)?
> 
> http://codereview.chromium.org/6901090/diff/1/src/snapshot-empty.cc
> File src/snapshot-empty.cc (right):
> 
> http://codereview.chromium.org/6901090/diff/1/src/snapshot-empty.cc#newcode36
> src/snapshot-empty.cc:36: 
> How about removing decompressed_ and instead adding raw_ (or something like
> that) to the input data, e.g.
> 
> const byte Snapshot::raw_data_[] = { 0 };
> const byte* Snapshot::data_ = NULL;

Forgot one thing - how about support for thins in the GYP file?

[mnaganov (inactive), 2011-04-29 12:07:58]

http://codereview.chromium.org/6901090/diff/1/SConstruct
File SConstruct (right):

http://codereview.chromium.org/6901090/diff/1/SConstruct#newcode343
SConstruct:343: },
On 2011/04/29 06:50:29, Søren Gjesse wrote:
> As this is Linux only I think this should be either inside the os:linux
section
> above or have a 'os:linux' inside. I think an os:linux here is the best
choice,
> and makes adding more platforms most logical.

Done.

http://codereview.chromium.org/6901090/diff/1/SConstruct#newcode488
SConstruct:488: 'CPPDEFINES':   ['COMPRESS_STARTUP_DATA_BZ2'],
On 2011/04/29 06:50:29, Søren Gjesse wrote:
> Ditto for the LIBS part.

Done.

http://codereview.chromium.org/6901090/diff/1/include/v8.h
File include/v8.h (right):

http://codereview.chromium.org/6901090/diff/1/include/v8.h#newcode2646
include/v8.h:2646: int id;
On 2011/04/29 06:50:29, Søren Gjesse wrote:
> Is this id actually required?

I think, no. As we don't expose to embedder what those ids mean, they make no
sense.

http://codereview.chromium.org/6901090/diff/1/include/v8.h#newcode2649
include/v8.h:2649: int decompressed_size;
On 2011/04/29 06:50:29, Søren Gjesse wrote:
> Should the compression algorithm be exposed here (enum or string)?

Makes sense. But not for every resource, as it is the same for all of them. I've
added another API function for returning it.

http://codereview.chromium.org/6901090/diff/1/include/v8.h#newcode2682
include/v8.h:2682: * The following 3 functions are to be used when V8 is built
with
On 2011/04/29 06:50:29, Søren Gjesse wrote:
> with -> with the

Done.

http://codereview.chromium.org/6901090/diff/1/samples/shell.cc
File samples/shell.cc (right):

http://codereview.chromium.org/6901090/diff/1/samples/shell.cc#newcode31
samples/shell.cc:31: #ifdef COMPRESS_STARTUP_DATA_BZ2
On 2011/04/29 06:50:29, Søren Gjesse wrote:
> Should this support be added to D8 as well? The process sample? For the
process
> sample maybe just add an #error.

I think, this doesn't make much sense for D8 either. I implemented decompression
support in shell just for demo purposes.

http://codereview.chromium.org/6901090/diff/1/src/mksnapshot.cc
File src/mksnapshot.cc (right):

http://codereview.chromium.org/6901090/diff/1/src/mksnapshot.cc#newcode130
src/mksnapshot.cc:130: bool Compress(Compressor* compressor) {
On 2011/04/29 06:50:29, Søren Gjesse wrote:
> ASSERT_EQ(-1, decompressed_size_)?

Done.

http://codereview.chromium.org/6901090/diff/1/src/snapshot-empty.cc
File src/snapshot-empty.cc (right):

http://codereview.chromium.org/6901090/diff/1/src/snapshot-empty.cc#newcode36
src/snapshot-empty.cc:36: 
On 2011/04/29 06:50:29, Søren Gjesse wrote:
> How about removing decompressed_ and instead adding raw_ (or something like
> that) to the input data, e.g.
> 
> const byte Snapshot::raw_data_[] = { 0 };
> const byte* Snapshot::data_ = NULL;

Good idea. "decompressed" is a too long word.
I ended up with using "raw" for naming decompressed data -- this feels more
natural than using "raw" for compressed.