gnuplot-beta

Re: How does ./configure interact with the demo directory?

[Hans-Bernhard B. <br...@ph...>]

On Tue, 20 Apr 2004, Petr Mikulik wrote:

> I prefer to keep "all.dem" as the main demo file.
> Then we don't need to bother with its generation by all possible
> config/makefile's.
> 
> The "local" one, you wish to ./configure, could be called
> all-local.dem, for example.

OTOH, it may be preferrable to have that generator, if we write one, write
both 'all.dem' and 'all-local.dem'.  The idea being that there should be
only one listing of all the demos in the source code, if possible.

The longer I look at this, the more it sounds like the GNU autogen tool
might be a good idea.

-- 
Hans-Bernhard Broeker (br...@ph...)
Even if all the snow were burnt, ashes would remain.

Re: How does ./configure interact with the demo directory?

[Hans-Bernhard B. <br...@ph...>]

On Wed, 21 Apr 2004, Lars Hecking wrote:

> Ethan Merritt writes:
> > While the dust is settling from the version 4 release,
> > I thought I'd have a go at adding a couple of new demos
> > and cleaning up the auto-generation of html demos.
> > 
> > Is there there some way to make the generation of
> > demos (e.g. their inclusion in 'all.dem' or 'make html')
> > depend on the options selected in ./configure?
> > 
> > For example, if you say
> > 	./configure --disable-filledboxes
> > then it should not try to exercise fillstyle.dem
>  
>  gnuplot does not have a proper test suite as such. A proper test suite,
>  GNU auto* style, would exercise gnuplot features, and print PASS/FAIL,
>  no graphics. I am not sure this can be done with gnuplot in the first place.

The "obvious" way would be a regression suite.  E.g. run each demo
through

	GNUTERM=postscript gnuplot foo.dem \
	  | filter_out_times_and_dates \
          | diff -q - foo_ref.ps 

and check the return status of diff to print PASS/FAIL.

>  Using the demos for 'make check' was my original attempt at something
>  that closely resembles a test suite.

I thought so.

> Hans-Bernhard Broeker writes:
> > On Mon, 19 Apr 2004, Ethan Merritt wrote:
> [...]
> > For html, you can quite certainly use automake conditionals to change the
> > list of html files being run.
> 
>  I believe this might unnecessarily complicate matters. The all-local.dem.in
>  approach can easily be done with configure, once the demo that require
>  specific configure options have been identified.

If we insert some more settings into configure.in, right.  config.status
itself doesn't do conditionals on translating .in files into something
else, it just does replacements. 

-- 
Hans-Bernhard Broeker (br...@ph...)
Even if all the snow were burnt, ashes would remain.

Re: How does ./configure interact with the demo directory?

[Hans-Bernhard B. <br...@ph...>]

On Wed, 21 Apr 2004, Petr Mikulik wrote:

> > As I recall, the objection was raised that Windows does not provide
> > any reasonable place to print such a message.
> 
> It would print it on screen.

No, it wouldn't.  Try it:
	
	wgnuplot c:\config.sys

(or whatever text file you happen to have lying around that should be
guaranteed to fail as a gnuplot script).

You'll find that nothing gets printed anywhere.  The problem is that
wgnuplot runs its own "xterm" equivalent (the text window), and that
vanishes as soon as the program exits, and isn't even made visible in
non-interactive execution.

> You mean stderr cannot be redirected to a file?

There *is* no stderr to be redirected, in the full GUI versions for 
Windows.  wgnuplot_pipe.exe might work, if run from a console window.

> The message could use "set print" file.

I don't think stderr goes there, and for a good reason, too.  We'ld need a
special 'set error' or so to redirect error messages to a file.

-- 
Hans-Bernhard Broeker (br...@ph...)
Even if all the snow were burnt, ashes would remain.

Re: How does ./configure interact with the demo directory?

[Petr M. <mi...@ph...>]

> 	wgnuplot c:\config.sys

Not like that, but
wgnuplot
gnuplot> load 'all.dem'
gnuplot> load "test-me.dem'

> There *is* no stderr to be redirected, in the full GUI versions for
> Windows.  wgnuplot_pipe.exe might work, if run from a console window.
>
> > The message could use "set print" file.
>
> I don't think stderr goes there, and for a good reason, too.  We'ld need a
> special 'set error' or so to redirect error messages to a file.

Or a command line option be a solution:
  wgnuplot --stderr=log.err

--
PM

Re: How does ./configure interact with the demo directory?

[Hans-Bernhard B. <br...@ph...>]

On Wed, 21 Apr 2004, Petr Mikulik wrote:

> > 	wgnuplot c:\config.sys
> 
> Not like that, but
> wgnuplot
> gnuplot> load 'all.dem'
> gnuplot> load "test-me.dem'

But that's not a way we can run 'make check'.  The checks are, and have to
be, non-interactive (even the 'make check-interactive' one is), as far as
the gnuplot engine is concerned, simply because they pass filenames to
gnuplot.  And as far as wgnuplot is concerned, that means all screen
messages end up in /dev/null.

-- 
Hans-Bernhard Broeker (br...@ph...)
Even if all the snow were burnt, ashes would remain.

Re: How does ./configure interact with the demo directory?

[Daniel J S. <dan...@ie...>]

Hans-Bernhard Broeker wrote:

>On Tue, 20 Apr 2004, Petr Mikulik wrote:
>
>  
>
>>I prefer to keep "all.dem" as the main demo file.
>>Then we don't need to bother with its generation by all possible
>>config/makefile's.
>>
>>The "local" one, you wish to ./configure, could be called
>>all-local.dem, for example.
>>    
>>
>
>OTOH, it may be preferrable to have that generator, if we write one, write
>both 'all.dem' and 'all-local.dem'.  The idea being that there should be
>only one listing of all the demos in the source code, if possible.
>

That's a good idea.  "All" should contain all.  However, then 'all.dem' 
will still not go all the way through.  Is there some small modification 
that can be made so that a syntax error does not cause gnuplot to stop 
when inside a file script?  Or perhaps the second level of files?  That 
is "load all.dem" calls "load" from inside the loaded file.  If there is 
a syntax error at that second level, perhaps just continue on?

Dan

Re: How does ./configure interact with the demo directory?

[Ethan M. <merritt@u.washington.edu>]

On Tuesday 20 April 2004 02:29 pm, Daniel J Sebald wrote:
>
> That's a good idea.  "All" should contain all.  However, then 'all.dem'
> will still not go all the way through.  

I don't so much care what "all.dem" contains specifically.
What I am concerned about is that "make check" should not fail
just because you disabled an option in ./configure.

> Is there some small modification 
> that can be made so that a syntax error does not cause gnuplot to stop
> when inside a file script?  

That is similar to an idea I had some time ago -
that when an option is disabled in ./configure there is still enough
residual code generated so that the syntax is legal:

#ifdef SOMETHING_FANCY
   lots of code
#else
   int_error(NO_CARET, "Your gnuplot does not support this option");
#endif

As I recall, the objection was raised that Windows does not provide
any reasonable place to print such a message.


-- 
Ethan A Merritt       merritt@u.washington.edu
Biomolecular Structure Center
Mailstop 357742
University of Washington, Seattle, WA 98195

Re: How does ./configure interact with the demo directory?

[Daniel J S. <dan...@ie...>]

Ethan Merritt wrote:

>On Tuesday 20 April 2004 02:29 pm, Daniel J Sebald wrote:
>  
>
>>That's a good idea.  "All" should contain all.  However, then 'all.dem'
>>will still not go all the way through.  
>>    
>>
>
>I don't so much care what "all.dem" contains specifically.
>What I am concerned about is that "make check" should not fail
>just because you disabled an option in ./configure.
>

Oh, I see.

>>Is there some small modification 
>>that can be made so that a syntax error does not cause gnuplot to stop
>>when inside a file script?  
>>    
>>
>
>That is similar to an idea I had some time ago -
>that when an option is disabled in ./configure there is still enough
>residual code generated so that the syntax is legal:
>

That is what I had in mind.

>#ifdef SOMETHING_FANCY
>   lots of code
>#else
>   int_error(NO_CARET, "Your gnuplot does not support this option");
>#endif
>
>As I recall, the objection was raised that Windows does not provide
>any reasonable place to print such a message.
>

Too bad.  It's nice to have gnuplot not act too differently depending 
upon platform.

Dan

Re: How does ./configure interact with the demo directory?

[Petr M. <mi...@ph...>]

> As I recall, the objection was raised that Windows does not provide
> any reasonable place to print such a message.

It would print it on screen.
You mean stderr cannot be redirected to a file?
The message could use "set print" file.

---
PM