Many error msgs for all user-documents in lout 3.38 (MS Vista + Cygwin)

[Volker Renneberg]

Hi there again!

Finally, I succeeded to compile lout 3.38 under Cygwin/Windows Vista.
However, I cannot compile any of the delivered lout documents (slides,
expert, user, design).

It gives a lot of messages of the form:
lout file "all":
69,1: width missing from 0i
69,1: /, | and & must be enclosed in double quotes

I have attached the compilation report for the slides-document.

I have set OSDOS to 1 and OSUNIX to 0. I have tried it with locale
turned on and off. I did not change any of the include-files.

Right now, I do not know where this problem comes from neither where to
look at.... Any ideas?

Cheers
Volker

[Ludovic Courtès]

Hello,

Post by Volker Renneberg
I have set OSDOS to 1 and OSUNIX to 0.

Did you try the opposite? Cygwin is supposed to provide a Unix-like
environment, so perhaps that would work better.

Thanks,
Ludo'.

[KHMan]

Post by Ludovic Courtès
Hello,

Post by Volker Renneberg
I have set OSDOS to 1 and OSUNIX to 0.

Did you try the opposite? Cygwin is supposed to provide a Unix-like
environment, so perhaps that would work better.

I can confirm that the errors also appear with the following
Makefile changes in WinXP/Cygwin:

PREFIX = /usr/local
USELOC = 0

That is, "OSUNIX = 1". Program was run after a "make install".
The same configuration and procedure worked fine in Ubuntu-8.04.

Anyone using Lout on Cygwin? I wanted to try it sometime ago but
haven't got around to it.

[KHMan]

Post by Volker Renneberg
Hi there again!
Finally, I succeeded to compile lout 3.38 under Cygwin/Windows Vista.
However, I cannot compile any of the delivered lout documents (slides,
expert, user, design).
69,1: width missing from 0i
69,1: /, | and & must be enclosed in double quotes
I have attached the compilation report for the slides-document.
I have set OSDOS to 1 and OSUNIX to 0. I have tried it with locale
turned on and off. I did not change any of the include-files.
Right now, I do not know where this problem comes from neither where to
look at.... Any ideas?

A follow up.

I guess the likely problem is in z19.c:223. I think there is a
small difference between the implementation of [s]scanf in newlibc
and gnulibc. From the ISO C standard draft n1256, it looks like
newlibc is correct, or maybe both are valid interpretations. The
use of sscanf in that part of the code looks like it really needs
to depend on pretty specific behaviour, fairly brittle usage.

I will patch my version later today and see if it runs properly on
cygwin with the fix.

[KHMan]

Post by Volker Renneberg
Finally, I succeeded to compile lout 3.38 under Cygwin/Windows Vista.
However, I cannot compile any of the delivered lout documents (slides,
expert, user, design).
69,1: width missing from 0i
69,1: /, | and & must be enclosed in double quotes
I have attached the compilation report for the slides-document.
I have set OSDOS to 1 and OSUNIX to 0. I have tried it with locale
turned on and off. I did not change any of the include-files.
Right now, I do not know where this problem comes from neither where to
look at.... Any ideas?

[Sorry for threading messages like this, I can never get back my
own e-mail from gmail, it's a known problem...]

Here's a stopgap measure, I got all the docs generated on
Cygwin/WinXP with the attached very ugly hack (z17.diff).
Basically I think the scanf theory holds.

I used the following changes to the Makefile:

PREFIX = /usr/local
USELOC = 0

And I have it installed to /usr/local before processing the
documents. Appears to go normally, the slides output is correct,
but I haven't tried to scrutinize the other docs yet.

I'm not really sure if newlib's sscanf is really at fault, because
its behaviour sounds wrong if we use "man scanf", but sounds right
if we use the ISO C draft standard n1256. So it seems all very
ambiguous...

[Jeff Kingston]

Could you explain what the issue is? What is wrong, or ambiguous,
about %f in a scanf format? It seems totally normal and unambiguous.

Jeff

Post by KHMan
I'm not really sure if newlib's sscanf is really at fault, because
its behaviour sounds wrong if we use "man scanf", but sounds right
if we use the ISO C draft standard n1256. So it seems all very
ambiguous...
- if( sscanf((char *) str, "%f", &num) != 1 )
+ if( sscanf((char *) str, "%31[0-9.]", &conv) != 1)

[KHMan]

Post by Jeff Kingston
Could you explain what the issue is? What is wrong, or ambiguous,
about %f in a scanf format? It seems totally normal and unambiguous.

It appears that in Cygwin, the sscanf works in a slightly
different manner. Possibly, it's a Cygwin newlib bug -- but I'll
study the newlib sources etc first before risking being flamed on
the Cygwin list...

When Lout sees "0i":

(1) On gnulibc, since 'i' is not a valid %f character, only '0' is
converted because conversion stops before the 'i'. "man scanf"
gives this interpretation.
(2) On newlib (Cygwin), sscanf grabs in a greedy manner, I think,
and sees '0i', and conversion fails.

Behaviour (2) seems likely wrong versus the man page. I'll look at
the newlib sources and the POSIX specifications later...

Post by Jeff Kingston

Post by KHMan
I'm not really sure if newlib's sscanf is really at fault, because
its behaviour sounds wrong if we use "man scanf", but sounds right
if we use the ISO C draft standard n1256. So it seems all very
ambiguous...
- if( sscanf((char *) str, "%f", &num) != 1 )
+ if( sscanf((char *) str, "%31[0-9.]", &conv) != 1)

[KHMan]

Post by Jeff Kingston
Could you explain what the issue is? What is wrong, or ambiguous,
about %f in a scanf format? It seems totally normal and unambiguous.

An update. As far as I can tell, it's a Cygwin bug, and I've
posted a query on the Cygwin list.

For the following test cases:

#include <stdio.h>
int main()
{
char *foo1 = "10i";
char *foo2 = "0i";
char *foo3 = "0.0i";
char *foo4 = "1.0i";
char *foo5 = "0.1i";
float f;
printf("%d ", sscanf(foo1, "%f", &f)); printf("%f\n", f);
printf("%d ", sscanf(foo2, "%f", &f)); printf("%f\n", f);
printf("%d ", sscanf(foo3, "%f", &f)); printf("%f\n", f);
printf("%d ", sscanf(foo4, "%f", &f)); printf("%f\n", f);
printf("%d ", sscanf(foo5, "%f", &f)); printf("%f\n", f);
}

On Cygwin:
$ ./test
1 10.000000
0 10.000000
1 0.000000
1 1.000000
1 0.100000

So it works fine for most things except '0'. Looks like a Cygwin
bug for sure. Once there is more information from the Cygwin
people, I will put up a brief wiki page on this for the Lout wiki.

Post by Jeff Kingston

Post by KHMan
I'm not really sure if newlib's sscanf is really at fault, because
its behaviour sounds wrong if we use "man scanf", but sounds right
if we use the ISO C draft standard n1256. So it seems all very
ambiguous...
- if( sscanf((char *) str, "%f", &num) != 1 )
+ if( sscanf((char *) str, "%31[0-9.]", &conv) != 1)

[Volker Renneberg]

Hi - thanks! This patch works fine for me (Vista and XP).

Cheers
Volker

Here's a stopgap measure, I got all the docs generated on Cygwin/WinXP
with the attached very ugly hack (z17.diff). Basically I think the
scanf theory holds. [.........................................]

[KHMan]

Post by Volker Renneberg
Hi - thanks! This patch works fine for me (Vista and XP).

Last update to the list. It's a bug and Redhat has patched newlib,
so the problem will disappear in a future version of the cygwin
runtime. No change to Lout mainline sources is necessary.

In the meantime, the ugly hack should be okay as a stopgap
measure, and I'll find some time to add a Lout-on-Cygwin wiki page.

Post by Volker Renneberg

Here's a stopgap measure, I got all the docs generated on Cygwin/WinXP
with the attached very ugly hack (z17.diff). Basically I think the
scanf theory holds. [.........................................]

[k***@cas.mcmaster.ca]

Hallo Volker! ;-)

Post by Volker Renneberg
Finally, I succeeded to compile lout 3.38 under Cygwin/Windows Vista.
However, I cannot compile any of the delivered lout documents (slides,
expert, user, design).

[...]

Post by Volker Renneberg
I have set OSDOS to 1 and OSUNIX to 0. I have tried it with locale
turned on and off. I did not change any of the include-files.

I think Cygwin counts as UNIX.


Hope this helps!


Wolfram