Re: Re: TeX run : 8

Re: Re: TeX run : 8

[Steffen Wolfrum]

Hi Patrick,



Patrick Gundlach <patrick@gundla.ch> wrote:

>  Hello again,
>
>  > I have tried your small file (the one you have sent me) an a vanilla
>  > tetex 3.0 and an older texlive (I had to dig out my iBook -- anybody
>  > wants to buy a used iBook?) and both were two runs.
>
>  Ah, not so fast, Patrick. I can reproduce your problem:
>
>  $ ls -lR
>  .:
>  total 4
>  drwxr-xr-x    4 pg       pg            136 Mar  2 20:27 PARTS
>  -rw-r--r--    1 pg       pg            117 Mar  2 20:28 TEST.tex
>
>  ./PARTS:
>  total 8
>  -rw-r--r--    1 pg       pg             40 Mar  2 20:27 ONE.tex
>  -rw-r--r--    1 pg       pg             40 Mar  2 20:27 TWO.tex
>
>  (so setup is TEST.tex in all capitals and ONE.tex and TWO.tex in
>  subdir, tetex 3.0). Now I run
>
>  texexec --pdf test.tex
>
>  (lowercase test.tex !)
>
>  and I get:
>
>  [...]
>
>   utility file analysis : another run needed
>                 TeX run : 8
>
>
>  [...]
>
>
>  Note that the standard OS X filesystem is case preserving, but not
>  case sensitive when accessing a file. So accessing test.tex would be
>  happy with a file called TEST.tex. It is possible to create a pure
>  case sensitive HFS+ filesystem, but it isn't necessary to reproduce
>  the problem. (It would probably avoid it, though).
>
>
>  Patrick



So I made a new setup:
new files/folders and every names lowercase (folders, files, references).

The first time I executed
texexec --pdf test.tex
it looked good: 2 runs!
Then I executed
texexec --pdf test.tex
once more and now it was 8 runs again!

Does this make sense?

Steffen

Re: TeX run : 8

[Patrick Gundlach]

Hello Steffen,


it is starting to make fun!

ok, clean setup, all lowercase. 8 runs. Now texutil --purgeall. 2
runs. purge again. 8 runs. purge again. 2 runs. No purge. 8 runs.
Purge. 8 runs. 2 runs. purge. 4 runs (!!!). purge. 2 runs. purge. 8
runs. 3 runs. purge. 5 runs. purge. 5 runs. purge. 8 runs.


-> 8, 2, 8, 2, 8, 8, 2, 4, 2, 8, 3, 5, 5, 8

I'll stop here. This looks a bit like the fibonacci numbers to me, but
the 4 must be an error.


Is it full moon?


Patrick

(no need to put me in cc anymore, I woke up)

Re: Re: TeX run : 8

[Hans Hagen]

Patrick Gundlach wrote:

> ok, clean setup, all lowercase. 8 runs. Now texutil --purgeall. 2
> runs. purge again. 8 runs. purge again. 2 runs. No purge. 8 runs.
> Purge. 8 runs. 2 runs. purge. 4 runs (!!!). purge. 2 runs. purge. 8
> runs. 3 runs. purge. 5 runs. purge. 5 runs. purge. 8 runs.
> 
> 
> -> 8, 2, 8, 2, 8, 8, 2, 4, 2, 8, 3, 5, 5, 8

can you patch texexec to provide some info about the file compare? (filesizes or 
so)

what puzzles me is that the tui/tuo does not change

Hans

-----------------------------------------------------------------
                                           Hans Hagen | PRAGMA ADE
               Ridderstraat 27 | 8061 GH Hasselt | The Netherlands
      tel: 038 477 53 69 | fax: 038 477 53 74 | www.pragma-ade.com
                                              | www.pragma-pod.nl
-----------------------------------------------------------------

Re: TeX run : 8

[Patrick Gundlach]

[-- Attachment #1: Type: text/plain, Size: 669 bytes --]

Hello Hans,


[...]

> can you patch texexec to provide some info about the file compare?
> (filesizes or so)

done, see attachment.


> what puzzles me is that the tui/tuo does not change

the .tuo file changes. 


% TeXUtil 9.0.0 - ConTeXt / PRAGMA ADE 1992-2004 / Files
%
% parts/one.tex (2)
% parts/two.tex (2)
% test (2)

% TeXUtil 9.0.0 - ConTeXt / PRAGMA ADE 1992-2004 / Files
%
% parts/two.tex (2)
% parts/one.tex (2)
% test (2)

% TeXUtil 9.0.0 - ConTeXt / PRAGMA ADE 1992-2004 / Files
%
% parts/two.tex (2)
% test (2)
% parts/one.tex (2)

% TeXUtil 9.0.0 - ConTeXt / PRAGMA ADE 1992-2004 / Files
%
% parts/two.tex (2)
% test (2)
% parts/one.tex (2)


Patrick

[-- Attachment #2: md5sum --]
[-- Type: application/octet-stream, Size: 4702 bytes --]

[-- Attachment #3: Type: text/plain, Size: 44 bytes --]


-- 
ConTeXt wiki: http://contextgarden.net

[-- Attachment #4: Type: text/plain, Size: 139 bytes --]

_______________________________________________
ntg-context mailing list
ntg-context@ntg.nl
http://www.ntg.nl/mailman/listinfo/ntg-context

Re: TeX run : 8

[Patrick Gundlach]

Hi,

I don't know enough of perl, perhaps the hash %Files should be sorted before
output (texutil.pl)?


sub FlushFiles
  { print TUO "%\n" . "% $Program / Files\n" . "%\n" ;
    foreach $File (keys %Files)
      { print TUO "% $File ($Files{$File})\n" }

Patrick
-- 
ConTeXt wiki: http://contextgarden.net

Re: Re: TeX run : 8

[Taco Hoekwater]

I think you've nailed it. Nice debugging!

Taco

Patrick Gundlach wrote:
> Hi,
> 
> I don't know enough of perl, perhaps the hash %Files should be sorted before
> output (texutil.pl)?
> 
> 
> sub FlushFiles
>   { print TUO "%\n" . "% $Program / Files\n" . "%\n" ;
>     foreach $File (keys %Files)
>       { print TUO "% $File ($Files{$File})\n" }
> 
> Patrick

Re: TeX run : 8

[Patrick Gundlach]

Hello again,

> I don't know enough of perl, perhaps the hash %Files should be sorted before
> output (texutil.pl)?

good point, Patrick :-)

this does the trick:

sub FlushFiles
  { print TUO "%\n" . "% $Program / Files\n" . "%\n" ;
    foreach $File (sort keys %Files)
      { print TUO "% $File ($Files{$File})\n" }
    print TUO "%\n" ;
    $NOfFiles = keys %Files ;

(in texutil.pl, see the _sort_ keys %Files)

But I don't know if this is a good solution.

Patrick
(no more "8 runs")
-- 
ConTeXt wiki: http://contextgarden.net

Re: Re: TeX run : 8

[Hans Hagen]

Patrick Gundlach wrote:
> Hello Hans,
> 
> 
> [...]
> 
> 
>>can you patch texexec to provide some info about the file compare?
>>(filesizes or so)
> 
> 
> done, see attachment.

now this is weird!

% parts/one.tex (2)
% parts/two.tex (2)
% test (2)

% parts/two.tex (2)
% parts/one.tex (2)
% test (2)

% parts/two.tex (2)
% test (2)
% parts/one.tex (2)

it looks like some flushing problem of write nodes, can you do the same with the 
tui file?

the code in context that does this is:

   \def\registerfileinfo[#1#2]#3%
     {\writestatus\m!systems{#1#2 file #3 at line \the\inputlineno}%
      \immediatewriteutility{f #1 {#3}}}

does it run ok when you define this as:

   \def\registerfileinfo[#1]#2{}

if so, it may be a problem in the tex binary (immediate writes not flushed ok)

Hans

-----------------------------------------------------------------
                                           Hans Hagen | PRAGMA ADE
               Ridderstraat 27 | 8061 GH Hasselt | The Netherlands
      tel: 038 477 53 69 | fax: 038 477 53 74 | www.pragma-ade.com
                                              | www.pragma-pod.nl
-----------------------------------------------------------------

Re: Re: TeX run : 8

[Hans Hagen]

Patrick Gundlach wrote:
> Hi,
> 
> I don't know enough of perl, perhaps the hash %Files should be sorted before
> output (texutil.pl)?
> 
> sub FlushFiles
>   { print TUO "%\n" . "% $Program / Files\n" . "%\n" ;
>     foreach $File (keys %Files)
>       { print TUO "% $File ($Files{$File})\n" }

That could be a way out, but before that i want you to check the order in the 
tui files.

[if it's indeed the perl hash, (maybe i should make it an array), then it looks 
like perl uses a different hash strategy each run and on each os]

Hans


-----------------------------------------------------------------
                                           Hans Hagen | PRAGMA ADE
               Ridderstraat 27 | 8061 GH Hasselt | The Netherlands
      tel: 038 477 53 69 | fax: 038 477 53 74 | www.pragma-ade.com
                                              | www.pragma-pod.nl
-----------------------------------------------------------------

Re: TeX run : 8

[Patrick Gundlach]

Hello Hans,


do you still want me to try this? 


> it looks like some flushing problem of write nodes, can you do the
> same with the tui file?
>
> the code in context that does this is:
>
>    \def\registerfileinfo[#1#2]#3%
>      {\writestatus\m!systems{#1#2 file #3 at line \the\inputlineno}%
>       \immediatewriteutility{f #1 {#3}}}
>
> does it run ok when you define this as:
>
>    \def\registerfileinfo[#1]#2{}
>
> if so, it may be a problem in the tex binary (immediate writes not flushed ok)

-- 
ConTeXt wiki: http://contextgarden.net

Re: Re: TeX run : 8

[Hans Hagen]

Patrick Gundlach wrote:

> do you still want me to try this? 

not if we assume that indeed perl hashes are organized differently per run

so, for the moment let stick to the texutil sort patch

thanks for testing; i'll post a new beta later

Hans


-----------------------------------------------------------------
                                           Hans Hagen | PRAGMA ADE
               Ridderstraat 27 | 8061 GH Hasselt | The Netherlands
      tel: 038 477 53 69 | fax: 038 477 53 74 | www.pragma-ade.com
                                              | www.pragma-pod.nl
-----------------------------------------------------------------

Re: Re: TeX run : 8

[Angus Lees]

At Thu, 03 Mar 2005 12:59:18 +0100, Hans Hagen wrote:
> [if it's indeed the perl hash, (maybe i should make it an array), then
> it looks like perl uses a different hash strategy each run and on each
> os]

Indeed, perl 5.8.1 introduced randomised hash functions to avoid the
"hash collision" attacks that were being published at the time.  You
can never assume hash traversal order will be constant across
different hashes.  See the discussion in the beginning of
`perldoc perl581delta' (and a followup in perl582delta).

-- 
 - Gus

Re[2]: Re: TeX run : 8

[Giuseppe Bilotta]

Thursday, March 3, 2005 Hans Hagen wrote:

> Patrick Gundlach wrote:
>> Hi,
>> 
>> I don't know enough of perl, perhaps the hash %Files should be sorted before
>> output (texutil.pl)?
>> 
>> sub FlushFiles
>>   { print TUO "%\n" . "% $Program / Files\n" . "%\n" ;
>>     foreach $File (keys %Files)
>>       { print TUO "% $File ($Files{$File})\n" }

> That could be a way out, but before that i want you to check the order in the
> tui files.

> [if it's indeed the perl hash, (maybe i should make it an array), then it looks
> like perl uses a different hash strategy each run and on each os]

http://search.cpan.org/~nwclark/perl-5.8.6/pod/perlfaq4.pod#Data:_Hashes_(Associative_Arrays)

-- 
Giuseppe "Oblomov" Bilotta

Re: Re: TeX run : 8

[Taco Hoekwater]

Giuseppe Bilotta wrote:
 >>
>>[if it's indeed the perl hash, (maybe i should make it an array), then it looks
>>like perl uses a different hash strategy each run and on each os]

The most relevant bit of documentation is this:

http://search.cpan.org/dist/perl/pod/perl581delta.pod#Hash_Randomisation

So yes, it is perl, and it is on purpose.

There might be more locations in ConTeXt's perl scripts
that need an additional 'sort'.

Greetings, Taco

Re: Re: TeX run : 8

[Peter Münster]

On Fri, 4 Mar 2005, Taco Hoekwater wrote:

> Giuseppe Bilotta wrote:
>  >>
> >>[if it's indeed the perl hash, (maybe i should make it an array), then it looks
> >>like perl uses a different hash strategy each run and on each os]
> 
> The most relevant bit of documentation is this:
> 
> http://search.cpan.org/dist/perl/pod/perl581delta.pod#Hash_Randomisation
> 
> So yes, it is perl, and it is on purpose.
> 
> There might be more locations in ConTeXt's perl scripts
> that need an additional 'sort'.

Hello,
for those, who don't want to update:
export PERL_HASH_SEED=0
in $HOME/.profile works very nicely.

Greetings, Peter

-- 
http://pmrb.free.fr/contact/

Re: TeX run : 8

[Patrick Gundlach]

Hello Peter,

> for those, who don't want to update:
> export PERL_HASH_SEED=0
> in $HOME/.profile works very nicely.

is this reliable?

Patrick
-- 
ConTeXt wiki: http://contextgarden.net

Re: Re: TeX run : 8

[Peter Münster]

On Tue, 8 Mar 2005, Patrick Gundlach wrote:

> > for those, who don't want to update:
> > export PERL_HASH_SEED=0
> > in $HOME/.profile works very nicely.
> 
> is this reliable?

It seems to me. At least on Linux: since a long time I've been used to get
always 8 runs with my procuct-files, and now I get only 1 or 2 runs.

Cheers, Peter

-- 
http://pmrb.free.fr/contact/

Re: TeX run : 8

[Patrick Gundlach]

Hello Adam,


[...]

> [And boy, it sure is faster than the setup I had on my PowerBook!]

Don't say that ;-). I just got a new PowerBook.

Patrick (and happy with it)
-- 
ConTeXt wiki: http://contextgarden.net

Re: TeX run : 8

[Patrick Gundlach]

Hello Hans,

[...]

> ah, i explicitly tried both lower and uppercase variants of the sample
> files, but windows does not care, i.e. all access deep down in the
> system is (afaik) cap. So i saw no difference here.

> so, what do you recomend, renaming the file to lowercase?

Well, my mail was sent too soon. The problems also arises without
the uppercase variant. So currently I doubt that it has to do with the
naming alone. 

Patrick

-- 
ConTeXt wiki: http://contextgarden.net

Re: TeX run : 8

[Patrick Gundlach]

Hello again,

> I have tried your small file (the one you have sent me) an a vanilla
> tetex 3.0 and an older texlive (I had to dig out my iBook -- anybody
> wants to buy a used iBook?) and both were two runs.

Ah, not so fast, Patrick. I can reproduce your problem:

$ ls -lR
.:
total 4
drwxr-xr-x    4 pg       pg            136 Mar  2 20:27 PARTS
-rw-r--r--    1 pg       pg            117 Mar  2 20:28 TEST.tex

./PARTS:
total 8
-rw-r--r--    1 pg       pg             40 Mar  2 20:27 ONE.tex
-rw-r--r--    1 pg       pg             40 Mar  2 20:27 TWO.tex

(so setup is TEST.tex in all capitals and ONE.tex and TWO.tex in
subdir, tetex 3.0). Now I run

texexec --pdf test.tex

(lowercase test.tex !)

and I get:

[...]

 utility file analysis : another run needed
               TeX run : 8


[...]


Note that the standard OS X filesystem is case preserving, but not
case sensitive when accessing a file. So accessing test.tex would be
happy with a file called TEST.tex. It is possible to create a pure
case sensitive HFS+ filesystem, but it isn't necessary to reproduce
the problem. (It would probably avoid it, though).


Patrick
-- 
ConTeXt wiki: http://contextgarden.net

Re: TeX run : 8

[Patrick Gundlach]

Hello Hans, Steffen and all the others involved,

> dunno, the mac is not updated yet and actually not sitting on my desk
> but in my library; best let patrick and/or adam test that since they
> normally have the latest (alpha/beta/current) versions

I have tried your small file (the one you have sent me) an a vanilla
tetex 3.0 and an older texlive (I had to dig out my iBook -- anybody
wants to buy a used iBook?) and both were two runs.

I do not use Gerben's tex, so we cannot really compare the results.

Patrick


(sorry for not reading this earlier)
-- 
ConTeXt wiki: http://contextgarden.net