On 11/10/2010 20:31, Adrian Thurston wrote:
Ya as far as Ragel code generation goes you shouldn't need to p, pe,
or data available in the write init context. If I remember right that
was due to a combination of Ruby variable scoping/declaration rules
and the typical usage pattern of Ragel+Ruby. The Ruby code generator
really is a special case for a number of reasons. What we need to a
Ragel and Ruby expert to sort it out. I'm not both of those things.
I'd take it on if I had the time.
There is also some strange stuff in the java code generator too. Java
doesn't have static array data. It makes functions that set the
values. But large arrays overflow the max function size so you can't
have large static arrays. To get around this the arrays are made using
functions (like the compiler does), but they are broken down into
smaller parts and concatenated. Ugh. I'm not sure if this would apply
to javascript as well.
I wrote a small test program and creating arrays with 1 and 10 million
elements worked, although at 10 million the VM was stuttering
(run-pause-run-pause), probably due to the GC not coping too well. How
large is large? In the case of Node.js there is another data type called
Buffer which is a fixed size byte array (sort of). This doesn't have a
literal notation but it does support the array [] operator so it might
be able to substitute for static arrays in code. It has the benefit of
being allocated outside of the runtime and so it doesn't pressurise the
GC as much. The downside is that it would require special IO handling
code to setup the arrays and it isn't portable.
It occured to me while writing this that I actually need to make at
least two different code generators for JavaScript. JavaScript can be
used for client and server programming (which is why I got into doing
this work in the first place) and the kind of code you want to generate
in each case is totally different. In the client case you want to
optimise for page load time first, so the state machine needs to be
expressed in the minimum possible code, even if it's slower to execute.
When generating the same machine for the server case you (probably) want
the normal approach of trading code size for performance.
Should be an interesting challenge.
Dominic
-Adrian
On 10-10-11 12:14 PM, Dominic Marks wrote:
Hey Adrian,
I haven't but I will do. Thank you.
I'm not sure if the RubyCodeGen::writeInit function is incorrect but
it's quite divergent from the CDCodeGen version. The only problem I had
doing the work, besides a few missed semicolons, was that the write init
directive was trying to initialise pe to the data length. In my testing
script the initialisation and execution stages are in separate functions
and this produced invalid JavaScript (data being undefined at that
point).
I assumed I had misunderstood something about the options that can be
given to write init and write exec but after that turned up a blank I
looked at CDCodeGen. I was surprised to find that it didn't include
similar code relating to the end directive. I made my writeInit function
more like CDCodeGen and my problem was resolved.
The code in question:
http://github.com/dominicmarks/ragel-js/commit/7a8e554f2829fbebe04a26837ebb5512314c376b#diff-0
The difference might be related to the concessions you mention, I didn't
look into it too much.
Thanks for Ragel, it's a great piece of software!
Dominic
On 11/10/2010 19:54, Adrian Thurston wrote:
Great, thanks for working on this!
Did you have a look at the java code generator? There were a few
concessions that had to be made in the Ruby code generator that might
not need to be made in JS.
-Adrian
On 10-10-11 04:07 AM, Dominic Marks wrote:
Hi all,
I've got the table code generation strategy working. I'm going to work
on building a set of RL files for testing purposes and then have a
crack
at writing a code generation backend that targets modern JavaScript
VMs
(e.g. produces JavaScript that best suits a JIT/tracer). If anyone
would
like to help with any part of this process, I'd be most grateful.
I've forked Ragel trunk into a github repo while I work on this. There
is a sample script (I translated the maillbox.rl file from the Ragel
distribution) which you can play with if you install Node.js (a
JavaScript application framework powered by Google's V8 VM). Note that
while the sample script uses Node.js for IO, but the code generated is
plain JavaScript. I look forward to being able to use Ragel in the
browser soon too.
Github:
http://github.com/dominicmarks/ragel-js
A diff against Ragel trunk is here:
http://goodforbusiness.co.uk/~dom/ragel-js-20101011.diff
<http://goodforbusiness.co.uk/%7Edom/ragel-js-20101011.diff>
The code is mostly a cut-paste-and-replace from Ruby at the moment.
I'll
be tidying it up over the next few weeks, as my time allows.
However, it
does at least run one Ragel state machine correctly already.
Mailbox.rl/Mailbox.js:
http://goodforbusiness.co.uk/~dom/mailbox.js
<http://goodforbusiness.co.uk/%7Edom/mailbox.js> (Requires Node.js).
Cheers,
Dom
_______________________________________________
ragel-users mailing list
[email protected]
http://www.complang.org/mailman/listinfo/ragel-users
_______________________________________________
ragel-users mailing list
[email protected]
http://www.complang.org/mailman/listinfo/ragel-users
_______________________________________________
ragel-users mailing list
[email protected]
http://www.complang.org/mailman/listinfo/ragel-users
_______________________________________________
ragel-users mailing list
[email protected]
http://www.complang.org/mailman/listinfo/ragel-users
_______________________________________________
ragel-users mailing list
[email protected]
http://www.complang.org/mailman/listinfo/ragel-users
