[MacRuby-devel] Proposal: splitting macrubyc UI from logic

[Laurent Sansonetti]

Hi Mark,

Sorry for the late reply.

Could you file at ticket and add a link to the changes on github? I will look into this once we release 0.10.

Thanks!
Laurent

On Mar 14, 2011, at 6:58 PM, Mark Rada wrote:

> On 2011-03-14, at 16:05, Laurent Sansonetti <lsansonetti at apple.com> wrote:
> 
>> Hi Mark,
>> 
>> As macrubyc's compilation logic is essentially spawning several command-line tools, I wonder if calling the logic directly from macruby_deploy is going to bring significant advantages, vs the complexity of splitting macrubyc.
>> 
> 
> The splitting macrubyc was a low hanging fruit; macrubyc was almost split already, so few changes were introduced. I don't think I introduced much complexity, and in turn some clutter was separated from the initialization process:
> 
> - calls to #die were replaced with calls to #raise
> - the option parser was moved out of the compiler class, now Compiler#initialize takes a hash of options and just unpacks it
> - most of the extra tool lookups (#locate) were moved to constants so they only have to be looked up once
> 
>> I think a better strategy would be to optimize what's slow in macrubyc (such as command-line options parsing),
> 
> I don't think it's the command line parsing, I thinks it's the spawning of new MacRuby processes which will have to JIT the compiler logic over and over again for each file. 
> 
> But I guess a lot of that can be mitigated by compiling the compiler when it becomes possible. 
> 
>> and better include the compilation strategy into Xcode (if possible).
>> 
> 
> That does sound like a much better idea for macruby_deploy. 
> 
> However, I am rarely using Xcode to work with MacRuby, and there are other places where calling the compiler directly will have benefit, such as a rake task or during gem installation. Perhaps I am speaking for a minority in these two cases
> 
> Sent from my iDevice
> 
>> Laurent
>> 
>> On Mar 12, 2011, at 5:40 PM, Mark Rada wrote:
>> 
>>> Hi,
>>> 
>>> I have completed a proof of concept patch for MacRuby where I have split the UI of macrubyc from the underlying logic so that tools like macruby_deploy can make use of the compiler without having to spawn a new macruby process for each file that needs to be compiled. This should also be beneficial for compiling gems and the standard library.
>>> 
>>> After having made this patch, I realized that there are still several places in the compiler where a new process is spawned to perform part of the compilation. I'm not really sure how much else can be lib-ified from the other required components. Overall there are still a few places that I know I can optimize without much work needed.
>>> 
>>> Right now, compile time for ruby files with about 100-200 lines of code is about 1(+/-0.1) seconds on my MBP. Spawning a new macruby process and processing the macrubyc options takes about 0.25 seconds; so I think the patch is still useful in the general case.
>>> 
>>> The code for the changes is located in my MacRuby fork on github: https://github.com/ferrous26/MacRuby/tree/libify-rubyc
>>> 
>>> Mark Rada
>>> mrada at marketcircle.com
>>> 
>>> _______________________________________________
>>> MacRuby-devel mailing list
>>> MacRuby-devel at lists.macosforge.org
>>> http://lists.macosforge.org/mailman/listinfo.cgi/macruby-devel
>> 
>> _______________________________________________
>> MacRuby-devel mailing list
>> MacRuby-devel at lists.macosforge.org
>> http://lists.macosforge.org/mailman/listinfo.cgi/macruby-devel
> 
> _______________________________________________
> MacRuby-devel mailing list
> MacRuby-devel at lists.macosforge.org
> http://lists.macosforge.org/mailman/listinfo.cgi/macruby-devel