Remove the PreCompile API and ScriptData.

Remove the PreCompile API and ScriptData.

The new compilation API (ScriptCompiler::Compile) can produce the same data, so
the separate precompilation phase is not needed. ScriptData is replaced by
ScriptCompiler::CachedData.

R=svenpanne@chromium.org
BUG=

Committed: https://code.google.com/p/v8/source/detail?r=20683

[marja, 2014-04-04 13:12:30]

svenpanne, PTAL.

(Note: this is just removing stuff from the API, not adding anything. Chromium
has already been updated to not use ScriptData.)

[Sven Panne, 2014-04-07 12:44:44]

https://codereview.chromium.org/225753004/diff/1/src/parser.cc
File src/parser.cc (right):

https://codereview.chromium.org/225753004/diff/1/src/parser.cc#newcode230
src/parser.cc:230: if (length % sizeof(unsigned) != 0) {
Why do we silently return here? Wouldn't an ASSERT be better?

https://codereview.chromium.org/225753004/diff/1/src/parser.cc#newcode234
src/parser.cc:234: // Copy the data to ensure it is properly aligned.
This comment doesn't really belong here.

https://codereview.chromium.org/225753004/diff/1/src/parser.cc#newcode237
src/parser.cc:237: if (reinterpret_cast<intptr_t>(data) % sizeof(unsigned) == 0)
{
I think restructuring this function makes things clearer, something along the
lines of:

    owns_store_ = reinterpret_cast<intptr_t>(data) % sizeof(unsigned) != 0;
    if (owns_store_) {
       <make an aligned copy and assign the start of it to data>
    }
    <wrap data/length into a vector>

https://codereview.chromium.org/225753004/diff/1/src/parser.h
File src/parser.h (right):

https://codereview.chromium.org/225753004/diff/1/src/parser.h#newcode85
src/parser.h:85: class ScriptDataImpl {
Rename this to simply ScriptData, or perhaps even something more descriptive.

https://codereview.chromium.org/225753004/diff/1/src/parser.h#newcode100
src/parser.h:100: static ScriptDataImpl* New(const char* data, int length);
I can't see the implementation of this function.

[marja, 2014-04-09 12:53:54]

svenpanne: Thanks for comments, could you take another look?

https://codereview.chromium.org/225753004/diff/1/src/parser.cc
File src/parser.cc (right):

https://codereview.chromium.org/225753004/diff/1/src/parser.cc#newcode230
src/parser.cc:230: if (length % sizeof(unsigned) != 0) {
On 2014/04/07 12:44:44, Sven Panne wrote:
> Why do we silently return here? Wouldn't an ASSERT be better?

This data is coming in via the API, maybe the original author (this func used to
be ScriptData::New) didn't want to use ASSERT to validate user input?

Nothing bad will happen if the user passes data which is obviously invalid; we
will (silently) ignore it and parsing will be slower.

Would you rather have an ASSERT here? What's the V8 policy wrt asserting if
input is invalid?

https://codereview.chromium.org/225753004/diff/1/src/parser.cc#newcode234
src/parser.cc:234: // Copy the data to ensure it is properly aligned.
On 2014/04/07 12:44:44, Sven Panne wrote:
> This comment doesn't really belong here.

Done.

https://codereview.chromium.org/225753004/diff/1/src/parser.cc#newcode237
src/parser.cc:237: if (reinterpret_cast<intptr_t>(data) % sizeof(unsigned) == 0)
{
On 2014/04/07 12:44:44, Sven Panne wrote:
> I think restructuring this function makes things clearer, something along the
> lines of:
> 
>     owns_store_ = reinterpret_cast<intptr_t>(data) % sizeof(unsigned) != 0;
>     if (owns_store_) {
>        <make an aligned copy and assign the start of it to data>
>     }
>     <wrap data/length into a vector>

Done.

https://codereview.chromium.org/225753004/diff/1/src/parser.h
File src/parser.h (right):

https://codereview.chromium.org/225753004/diff/1/src/parser.h#newcode85
src/parser.h:85: class ScriptDataImpl {
On 2014/04/07 12:44:44, Sven Panne wrote:
> Rename this to simply ScriptData, or perhaps even something more descriptive.

Done.

https://codereview.chromium.org/225753004/diff/1/src/parser.h#newcode100
src/parser.h:100: static ScriptDataImpl* New(const char* data, int length);
On 2014/04/07 12:44:44, Sven Panne wrote:
> I can't see the implementation of this function.

Oops. This doesn't exist, ScriptDataImpl(const char data*, int length) does what
this was intended to do.

[Sven Panne, 2014-04-11 08:18:22]

https://codereview.chromium.org/225753004/diff/1/src/parser.cc
File src/parser.cc (right):

https://codereview.chromium.org/225753004/diff/1/src/parser.cc#newcode230
src/parser.cc:230: if (length % sizeof(unsigned) != 0) {
I think that silently (not) doing something is quite bad and will bite us sooner
or later. The corresponding API function has a way to communicate failure by
returning an empty handle, so this is what should happen somehow. The question
is how to communicate this through the layers, failing in the constructor
without having exceptions is ugly, so one could e.g. use a factory method
instead, but I'll leave the details up to you. :-)

[marja, 2014-04-11 09:04:33]

Offline discussion -> I'll do a follow-up change where we fail the compilation
if the cached data is invalid.

I added a function ScriptData::New and removed the ctor(const char*, int), to
make that change easier.

[Sven Panne, 2014-04-11 11:20:29]

lgtm

[marja, 2014-04-11 11:45:00]

Committed patchset #7 manually as r20683 (presubmit successful).