On 14 déc. 04, at 09:20, Andreas Raab wrote:

>> The new compiler sounds like a step forward by many measures.  The 
>> performance issue needs to be quantified more.  When you say it is 
>> slower, is it 5% slower? That would be tolerable.  If it is 20% 
>> slower it might not be tolerable.
>
> Running a quick benchmark "[Object compileAll] timeRun" gives:
>
> Old compiler:  1794 1749 1723
> New compiler:  4417 4384 4423
>
> So it's about a factor of 2.5 where this might be a little skewed 
> since I just loaded the closure compiler (which is a little more 
> complex than the non-closure compiling version) and switched between 
> the two. Hm... this is significantly more than I expected... and if we 
> actually cut out the time it takes for retrieving the source code by, 
> e.g., doing:
>
> code := Object selectors asArray
>    collect:[:sel| Object sourceCodeAt: sel].
> compiler := Object compilerClass.
> [code do:[:src|
>    compiler new
>        compile: src
>        in: Object
>        notifying: nil
>        ifFail:[]]
> ] timeToRun.
>
> We're almost precisely at a factor of 3 or so (spending ~50% of the 
> time in SmaCCScanner>>next btw).
>
>> If it breaks eToys, I would think you will need to get
>> an estimate from them for what is required to deal with this.
>
> Well, the real issue which I'd like to see fixed is error reporting. I 
> have found that SmaCC parsers are notoriously bad at it - except from 
> "token not expected" it doesn't tell you anything about what is wrong 
> (and it doesn't even tell you the token - so if you don't know how the 
> next characters compose you are at a total loss). For example, 
> consider the following erronous expressions:
>
>            SmaCC                  Squeak
> a := .      Token not expected     Expression expected
> 3 foo: +5   Token not expected     Argument expected
> [:foo ^3+4] Token not expected     Vertical bar expected
> [^42        Token not expected     Period or right bracket expected
>
>
> For a newbie this is deeply problematic. It's hard enough to get this 
> stuff right even if the compiler tells you what it expects - but if 
> you only know that there is "something wrong here" (perhaps a missing 
> \item? ;-))) it's almost impossible to find out.

Yes this is a problem.
So what do we do.