May 06, 2010
Walter Bright wrote:
> Walter Bright wrote:
>> Alex Makhotin wrote:
>>> It takes ~40 seconds 50% load on the dual core processor(CentOS 5.3 kernel 2.6.32.4), to get the actual error messages about the undefined identifier.
>>
>> Definitely there's a problem.
> 
> The problem is the spell checker is O(n*n) on the number of characters in the undefined identifier.

Is there a way to disable it?

-- 
Alex Makhotin,
the founder of BITPROX,
http://bitprox.com
May 06, 2010
Alex Makhotin wrote:
> Walter Bright wrote:
>> Walter Bright wrote:
>>> Alex Makhotin wrote:
>>>> It takes ~40 seconds 50% load on the dual core processor(CentOS 5.3 kernel 2.6.32.4), to get the actual error messages about the undefined identifier.
>>>
>>> Definitely there's a problem.
>>
>> The problem is the spell checker is O(n*n) on the number of characters in the undefined identifier.
> 
> Is there a way to disable it?
> 

Currently, no.
May 06, 2010
On Wed, 05 May 2010 23:45:50 -0400, Walter Bright <newshound1@digitalmars.com> wrote:

> Walter Bright wrote:
>> Alex Makhotin wrote:
>>> It takes ~40 seconds 50% load on the dual core processor(CentOS 5.3 kernel 2.6.32.4), to get the actual error messages about the undefined identifier.
>>  Definitely there's a problem.
>
> The problem is the spell checker is O(n*n) on the number of characters in the undefined identifier.

That can't be it.  The identifier shown by Alex is only 33 characters.  O(n^2) is not that slow, especially for smaller variables.  There must be other factors you're not considering...

-Steve
May 06, 2010
Hello Walter,

> Walter Bright wrote:
> 
>> Alex Makhotin wrote:
>> 
>>> It takes ~40 seconds 50% load on the dual core processor(CentOS 5.3
>>> kernel 2.6.32.4), to get the actual error messages about the
>>> undefined identifier.
>>> 
>> Definitely there's a problem.
>> 
> The problem is the spell checker is O(n*n) on the number of characters
> in the undefined identifier.
> 

How about switch algos for long identifiers: you could bucket the knows by length and compare histograms on things of similar length. Or maybe just turn it off for long names. 

-- 
... <IXOYE><



May 06, 2010
On 2010-05-05 23:45:50 -0400, Walter Bright <newshound1@digitalmars.com> said:

> Walter Bright wrote:
>> Alex Makhotin wrote:
>>> It takes ~40 seconds 50% load on the dual core processor(CentOS 5.3 kernel 2.6.32.4), to get the actual error messages about the undefined identifier.
>> 
>> Definitely there's a problem.
> 
> The problem is the spell checker is O(n*n) on the number of characters in the undefined identifier.

That's an algorithm that can't scale then.

Checking the Levenshtein distance for each known identifier within a small difference in length would be a better idea. (Clang is said to use the Levenshtein distance, it probably does something of the sort.)

http://en.wikipedia.org/wiki/Levenshtein_distance

-- 
Michel Fortin
michel.fortin@michelf.com
http://michelf.com/

May 06, 2010
Steven Schveighoffer, el  6 de mayo a las 07:17 me escribiste:
> On Wed, 05 May 2010 23:45:50 -0400, Walter Bright <newshound1@digitalmars.com> wrote:
> 
> >Walter Bright wrote:
> >>Alex Makhotin wrote:
> >>>It takes ~40 seconds 50% load on the dual core processor(CentOS 5.3 kernel 2.6.32.4), to get the actual error messages about the undefined identifier.
> >> Definitely there's a problem.
> >
> >The problem is the spell checker is O(n*n) on the number of
> >characters in the undefined identifier.
> 
> That can't be it.  The identifier shown by Alex is only 33 characters.  O(n^2) is not that slow, especially for smaller variables.  There must be other factors you're not considering...

Run a profiler.

-- 
Leandro Lucarella (AKA luca)                     http://llucax.com.ar/
----------------------------------------------------------------------
GPG Key: 5F5A8D05 (F8CD F9A7 BF00 5431 4145  104C 949E BFB6 5F5A 8D05)
----------------------------------------------------------------------
No existiría el sonido del mar si faltara en la vida oreja y caracol.
	-- Ricardo Vaporeso. Cosquín, 1908.
May 06, 2010
Steven Schveighoffer wrote:
> That can't be it.  The identifier shown by Alex is only 33 characters.  O(n^2) is not that slow, especially for smaller variables.  There must be other factors you're not considering...

I recompiled dmd with the profiler (-gt switch) which confirmed it.
May 06, 2010
On Thu, 06 May 2010 17:07:12 -0400, Walter Bright <newshound1@digitalmars.com> wrote:

> Steven Schveighoffer wrote:
>> That can't be it.  The identifier shown by Alex is only 33 characters.  O(n^2) is not that slow, especially for smaller variables.  There must be other factors you're not considering...
>
> I recompiled dmd with the profiler (-gt switch) which confirmed it.

So a single unknown symbol (from Alex's example) which can be checked against each existing symbol in O(n^2) time, takes 40 seconds on a decent CPU?  How many other symbols are there?  33^2 == 1089, if there are 10000 symbols, that's 10 million iterations, that shouldn't take 40 seconds to run, should it?  Are there more symbols to compare against?  Do you use heuristics to prune the search?  For example, if the max distance is 2, and the difference in length between two strings is >2, you should be able to return immediately.

-Steve
May 06, 2010
Steven Schveighoffer wrote:
> On Thu, 06 May 2010 17:07:12 -0400, Walter Bright <newshound1@digitalmars.com> wrote:
> 
>> Steven Schveighoffer wrote:
>>> That can't be it.  The identifier shown by Alex is only 33 characters.  O(n^2) is not that slow, especially for smaller variables.  There must be other factors you're not considering...
>>
>> I recompiled dmd with the profiler (-gt switch) which confirmed it.
> 
> So a single unknown symbol (from Alex's example) which can be checked against each existing symbol in O(n^2) time, takes 40 seconds on a decent CPU?  How many other symbols are there?  33^2 == 1089, if there are 10000 symbols, that's 10 million iterations, that shouldn't take 40 seconds to run, should it?  Are there more symbols to compare against?  Do you use heuristics to prune the search?  For example, if the max distance is 2, and the difference in length between two strings is >2, you should be able to return immediately.

Check out the profiler output. It's clearly the vast number of calls to the symbol lookup, not the time spent in each call.

-----------------------------------------
  Num          Tree        Func        Per
  Calls        Time        Time        Call

50409318   632285778   145858160           2     Dsymbol *syscall ScopeDsymbol::search(Loc ,Identifier *,int )
50411264   131394915   106356855           2     void **syscall StringTable::search(char const *,unsigned )
50409329   341960075   105532978           2     Dsymbol *syscall DsymbolTable::lookup(Identifier *)
50409329   236427096   105037393           2     StringValue *syscall StringTable::lookup(char const *,unsigned )
12602340   613890619    67393753           5     Dsymbol *syscall Scope::search(Loc ,Identifier *,Dsymbol **)
12602178   693915197    66918360           5     void *cdecl scope_search_fp(void *,char const *)
25204505   461352920    52529164           2     Dsymbol *syscall Module::search(Loc ,Identifier *,int )
50412137    25038474    25038474           0     unsigned cdecl Dchar::calcHash(char const *,unsigned )
   3520  1428323068    20349375        5781     void *cdecl spellerX(char const *,void *cdecl (*)(void *,char const *),void *,char const *,int )
12602664     6811916     6811916           0     syscall Identifier::Identifier(char const *,int )
12602178     6299089     6299089           0     void cdecl Module::clearCache()
12602183     6151175     6151175           0     Module *syscall Module::isModule()
   1600       11329        4261           2     StringValue *syscall StringTable::update(char const *,unsigned )

-----------------------------------------
May 07, 2010
On 6-5-2010 22:37, Michel Fortin wrote:
> On 2010-05-05 23:45:50 -0400, Walter Bright <newshound1@digitalmars.com> said:
> 
>> Walter Bright wrote:
>>> Alex Makhotin wrote:
>>>> It takes ~40 seconds 50% load on the dual core processor(CentOS 5.3 kernel 2.6.32.4), to get the actual error messages about the undefined identifier.
>>>
>>> Definitely there's a problem.
>>
>> The problem is the spell checker is O(n*n) on the number of characters
>> in the undefined identifier.
> 
> That's an algorithm that can't scale then.
> 
> Checking the Levenshtein distance for each known identifier within a small difference in length would be a better idea. (Clang is said to use the Levenshtein distance, it probably does something of the sort.)
> 
> http://en.wikipedia.org/wiki/Levenshtein_distance
> 
and especially this line:

# If we are only interested in the distance if it is smaller than a threshold k, then it suffices to compute a diagonal stripe of width 2k+1 in the matrix. In this way, the algorithm can be run in O(kl) time, where l is the length of the shortest string.