The purpose of this change is to avoid the need to index register
variables storing match results of sequences whose elements are labeled.
The indexing happened when match results of labeled elements were passed
to action/predicate functions.
In order to avoid indexing, the register allocator needs to ensure that
registers storing match results of any labeled sequence elements are
still "alive" after finishing parsing of the sequence. They should not
be used to store anything else at least until code of all actions and
predicates that can see the labels is executed. This requires that the
|allocateRegisters| pass has the knowledge of scoping. Because that
knowledge was already implicitly embedded in the |coputeParams| pass,
the logical step to prevent duplication was to merge it with the
|allocateRegisters| pass. This is what this commit does.
As a part of the merge the tests of both passes were largely refactored.
This is both to accomodate the merge and to make the tests in sync with
the code again (the tests became a bit out-of-sync during the last few
commits -- they tested more than was needed).
The speed/size impact is slightly positive:
Speed impact
------------
Before: 849.86 kB/s
After: 858.16 kB/s
Difference: 0.97%
Size impact
-----------
Before: 876618 b
After: 875602 b
Difference: -0.12%
(Measured by /tools/impact with Node.js v0.6.18 on x86_64 GNU/Linux.)
This commit changes the model underlying parser variables used to store
match results and parse positions. Until now they were treated as a
stack, now they are thought of as registers. The actual behavior does
not change (yet), only the terminology.
More specifically, this commit:
* Changes parser variable names from |result0|, |result1|, etc. to
|r0|, |r1|, etc.
* Changes various internal names and comments to match the new model.
* Renames the |computeVarIndices| pass to |allocateRegisters|.
This commit replaces all variable name computations in |computeVarNames|
and |computeParams| passes by computations of indices. The actual names
are computed later in the |generateCode| pass.
This change makes the code generator the only place that deals with the
actual variable names, making them easier to change for example.
The code generator code seems bit more complicated after the change, but
this complexity will pay off (and mostly disappear) later.
This has two main benefits:
1. The knowledge about scoping params in at one designated place,
making all future adjustments in this area easier.
2. Action-related code does not handle sequences specially anymore.
Such knowledge/behavior doesn't belong there.
Before this change, knowledge about variable names was spread between
the |computeStackDepths| pass and the code emitter code. For example,
the fact that the |&...| expression needs one variable to store a
position was represented in both places.
This changes consolidates that knowledge and introduces a new
|computeVarNames| pass. This pass replaces old |computeStackDepths|
pass, does all computations realted to variable names and stores the
results in the AST. Note that some knowledge about variables
(inevitably) remained in emitter code templates.
Beside DRYing things up, this change simplifies the emitter
significantly. By storing variable names in the AST it also allows
introduction of a pass that will identify parameters passed to actions
using proper symbol tables. Right now, this is done in a hackish way
directly in the emitter, which won't work well with changes planned in
GH-69.
Before this commit, variables for saving match results and parse
positions in generated parsers were not used efficiently. Each rule
basically used its own variable(s) for storing the data, with names
generated sequentially during code emitting. There was no reuse of
variables and a lot of unnecessary assignments between them.
It is easy to see that both match results and parse positions can
actually be stored on a stack that grows as the parser walks deeper in
the grammar tree and shrinks as it returns. Moreover, if one creates a
new stack for each rule the parser enters, its maximum depth can be
computed statically from the grammar. This allows us to implement the
stack not as an array, but as a set of numbered variables in each
function that handles parsing of a grammar rule, avoiding potentially
slow array accesses.
This commit implements the idea from the previous paragraph, using
separate stack for match results and for parse positions. As a result,
defined variables are reused and unnecessary copying avoided.
Speed implications
------------------
This change speeds up the benchmark suite execution by 2.14%.
Detailed results (benchmark suite totals as reported by "jake benchmark"
on Node.js 0.4.8):
-----------------------------------
Test # Before After
-----------------------------------
1 129.01 kB/s 131.98 kB/s
2 129.39 kB/s 130.13 kB/s
3 128.63 kB/s 132.57 kB/s
4 127.53 kB/s 129.82 kB/s
5 127.98 kB/s 131.80 kB/s
-----------------------------------
Average 128.51 kB/s 131.26 kB/s
-----------------------------------
Size implications
-----------------
This change makes a sample of generated parsers 8.60% smaller:
Before:
$ wc -c src/parser.js examples/*.js
110867 src/parser.js
13886 examples/arithmetics.js
450125 examples/css.js
632390 examples/javascript.js
61365 examples/json.js
1268633 total
After:
$ wc -c src/parser.js examples/*.js
99597 src/parser.js
13077 examples/arithmetics.js
399893 examples/css.js
592044 examples/javascript.js
54797 examples/json.js
1159408 total
The source code is now in the src directory. The library needs to be
built using "rake", which creates the lib/peg.js file by combining the
source files.