FvwmScript - WriteToFile seems not working
Hi! I'm working on a composite configurator with FvwmScript and getting an issue with WriteToFile command. It only writes '#end' into the file. First I thought I've done something wrong but I do the same as in FvwmScript-BaseConfig. So I started a test with FvwmScript-BaseConfig and the same happens. Used FVWM is 2.6.5. This happens under Fedora 19, too. My next thought was something has changed in WriteToFile source since creation of FvwmScript-BaseConfig - 2007-08-07. So I compared the code from 2.5.22 and CVS but nothing has changed. So, what could be the problem? - Thomas -
Re: FvwmScript - WriteToFile seems not working
Thomas Funk writes: > Hi! > > I'm working on a composite configurator with FvwmScript and getting an > issue with WriteToFile command. It only writes '#end' into the file. > > First I thought I've done something wrong but I do the same as in > FvwmScript-BaseConfig. So I started a test with FvwmScript-BaseConfig > and the same happens. Used FVWM is 2.6.5. This happens under Fedora 19, > too. > > My next thought was something has changed in WriteToFile source since > creation of FvwmScript-BaseConfig - 2007-08-07. So I compared the code > from 2.5.22 and CVS but nothing has changed. > > So, what could be the problem? We'd probably have to see the script but you could always resort to adding printfs in Instructions.c. Start with determining if it gets NbArg right. Just seeing #end is an indication that it doesn't see the args after the filename. -- Dan Espen
Re: FvwmScript - WriteToFile seems not working
Le Wed, 11 Dec 2013 20:29:22 +0100, Thomas Funk a écrit : > Hi! > > I'm working on a composite configurator with FvwmScript and getting an > issue with WriteToFile command. It only writes '#end' into the file. I get the same issue yesterday. I use a Piperead for now. The script is here: http://sourceforge.net/p/fvwm-crystal/code/HEAD/tree/fvwm/scripts/PrefVars/PrefVars The commented out WriteToFile is into Widget 23. This is with fvwm-2.6.6 from cvs and gentoo. Dominique > > First I thought I've done something wrong but I do the same as in > FvwmScript-BaseConfig. So I started a test with FvwmScript-BaseConfig > and the same happens. Used FVWM is 2.6.5. This happens under Fedora > 19, too. > > My next thought was something has changed in WriteToFile source since > creation of FvwmScript-BaseConfig - 2007-08-07. So I compared the > code from 2.5.22 and CVS but nothing has changed. > > So, what could be the problem? > > - Thomas - >
Re: FvwmScript - WriteToFile seems not working
Dan Espen wrote:
> Thomas Funk writes:
>
>> Hi!
>>
>> I'm working on a composite configurator with FvwmScript and getting an
>> issue with WriteToFile command. It only writes '#end' into the file.
>>
>> First I thought I've done something wrong but I do the same as in
>> FvwmScript-BaseConfig. So I started a test with FvwmScript-BaseConfig
>> and the same happens. Used FVWM is 2.6.5. This happens under Fedora 19,
>> too.
>>
>> My next thought was something has changed in WriteToFile source since
>> creation of FvwmScript-BaseConfig - 2007-08-07. So I compared the code
>> from 2.5.22 and CVS but nothing has changed.
>>
>> So, what could be the problem?
>
> We'd probably have to see the script but you could always resort to
> adding printfs in Instructions.c.
>
> Start with determining if it gets NbArg right.
> Just seeing #end is an indication that it doesn't
> see the args after the filename.
I have checked different VMs with different FVWM versions with
FvwmScript-BaseConfig:
Debian 4.x (2007) with FVWM 2.5.22 - ok, creates full config
Debian 6.x (2011) with FVWM 2.6.0 - ok, creates full config
Debian 7.1 (2013) with FVWM 2.6.6 - ok, creates full config
Debian 7.1 (2013) with FVWM 2.6.5 - ok, creates full config
Debian 8 (testing, 2013) with FVWM 2.6.5 - nok, creates '#end'
Fedora 19 (2013) with FVWM 2.6.5 - nok, creates '#end'
Dominique Michel wrote:
> I get the same issue yesterday.
> This is with fvwm-2.6.6 from cvs and gentoo.
So, it seems it's not a script problem.
Therefore I've starting your suggestion with 'printfs' in instructions.c
under my Fedora VM. I used this FvwmScript with one WriteToFile:
#WindowTitle {Test WriteToFile}
#WindowSize 470 415# Taille
#
#Init
#Begin
#Set $userDir = (GetOutput {echo "$FVWM_USERDIR"} 1 -1)
#Set $configPath = $userDir{/bla}
#
#WriteToFile $configPath {This is a test}
#Quit
#End
As I didn't knew what you meant with 'printfs' I used printf first and
hoped that output went into .xsession-errors. No output appeared but
WriteToFile worked. So I removed printf and compiled code suddenly
worked!
Same code, different behaviors...
Therefore I did the same on my Debian 8 system and the compiled version
of FvwmScript works also.
Under Debian 8 and Fedora 19 I have distribution packages installed.
All others are self compiled.
Could it be a problem of code optimization? Because the FvwmScript
executables has huge different sizes:
package: 380K
compiled: 1,5M
The only thing that not fit in this theory is Dominiques issue because
his code is self compiled, too ...
- Thomas -
Re: FvwmScript - WriteToFile seems not working
Thomas Funk writes:
> Dan Espen wrote:
>> Thomas Funk writes:
>>
>>> Hi!
>>>
>>> I'm working on a composite configurator with FvwmScript and getting an
>>> issue with WriteToFile command. It only writes '#end' into the file.
>>>
>>> First I thought I've done something wrong but I do the same as in
>>> FvwmScript-BaseConfig. So I started a test with FvwmScript-BaseConfig
>>> and the same happens. Used FVWM is 2.6.5. This happens under Fedora 19,
>>> too.
>>>
>>> My next thought was something has changed in WriteToFile source since
>>> creation of FvwmScript-BaseConfig - 2007-08-07. So I compared the code
>>> from 2.5.22 and CVS but nothing has changed.
>>>
>>> So, what could be the problem?
>>
>> We'd probably have to see the script but you could always resort to
>> adding printfs in Instructions.c.
>>
>> Start with determining if it gets NbArg right.
>> Just seeing #end is an indication that it doesn't
>> see the args after the filename.
>
> I have checked different VMs with different FVWM versions with
> FvwmScript-BaseConfig:
>
> Debian 4.x (2007) with FVWM 2.5.22 - ok, creates full config
> Debian 6.x (2011) with FVWM 2.6.0 - ok, creates full config
> Debian 7.1 (2013) with FVWM 2.6.6 - ok, creates full config
> Debian 7.1 (2013) with FVWM 2.6.5 - ok, creates full config
> Debian 8 (testing, 2013) with FVWM 2.6.5 - nok, creates '#end'
> Fedora 19 (2013) with FVWM 2.6.5 - nok, creates '#end'
Good stuff!
> Dominique Michel wrote:
>> I get the same issue yesterday.
>> This is with fvwm-2.6.6 from cvs and gentoo.
>
> So, it seems it's not a script problem.
>
> Therefore I've starting your suggestion with 'printfs' in instructions.c
> under my Fedora VM. I used this FvwmScript with one WriteToFile:
>
> #WindowTitle {Test WriteToFile}
> #WindowSize 470 415# Taille
> #
> #Init
> #Begin
> #Set $userDir = (GetOutput {echo "$FVWM_USERDIR"} 1 -1)
> #Set $configPath = $userDir{/bla}
> #
> #WriteToFile $configPath {This is a test}
> #Quit
> #End
>
> As I didn't knew what you meant with 'printfs' I used printf first and
> hoped that output went into .xsession-errors. No output appeared but
> WriteToFile worked. So I removed printf and compiled code suddenly
> worked!
>
> Same code, different behaviors...
>
> Therefore I did the same on my Debian 8 system and the compiled version
> of FvwmScript works also.
>
> Under Debian 8 and Fedora 19 I have distribution packages installed.
> All others are self compiled.
>
> Could it be a problem of code optimization? Because the FvwmScript
> executables has huge different sizes:
> package: 380K
> compiled: 1,5M
Hmm, I wonder what flags we need to use to get near that size?
> The only thing that not fit in this theory is Dominiques issue because
> his code is self compiled, too ...
I should have said fprintf(stderr);
I put a few in and observed expected behavior.
One near the end of CalcArg:
fprintf(stderr,"calc arg value of %s\n",TmpStr);
One right at the start of WriteToFile:
fprintf(stderr,"number of args %d\n",NbArg);
What you say above sounds like a very possible optimizer/compiler
problem. Probably the distro flags haven't changed so we might
need to know the flags and the specific compiler version.
BTW, I'm FC19 but I always build my own executables.
--
Dan Espen
Re: FvwmScript - WriteToFile seems not working
Le Sat, 14 Dec 2013 00:34:38 +0100,
Thomas Funk a écrit :
> Dan Espen wrote:
> > Thomas Funk writes:
> >
> >> Hi!
> >>
> >> I'm working on a composite configurator with FvwmScript and
> >> getting an issue with WriteToFile command. It only writes '#end'
> >> into the file.
> >>
> >> First I thought I've done something wrong but I do the same as in
> >> FvwmScript-BaseConfig. So I started a test with
> >> FvwmScript-BaseConfig and the same happens. Used FVWM is 2.6.5.
> >> This happens under Fedora 19, too.
> >>
> >> My next thought was something has changed in WriteToFile source
> >> since creation of FvwmScript-BaseConfig - 2007-08-07. So I
> >> compared the code from 2.5.22 and CVS but nothing has changed.
> >>
> >> So, what could be the problem?
> >
> > We'd probably have to see the script but you could always resort to
> > adding printfs in Instructions.c.
> >
> > Start with determining if it gets NbArg right.
> > Just seeing #end is an indication that it doesn't
> > see the args after the filename.
>
> I have checked different VMs with different FVWM versions with
> FvwmScript-BaseConfig:
>
> Debian 4.x (2007) with FVWM 2.5.22 - ok, creates full config
> Debian 6.x (2011) with FVWM 2.6.0 - ok, creates full config
> Debian 7.1 (2013) with FVWM 2.6.6 - ok, creates full config
> Debian 7.1 (2013) with FVWM 2.6.5 - ok, creates full config
> Debian 8 (testing, 2013) with FVWM 2.6.5 - nok, creates '#end'
> Fedora 19 (2013) with FVWM 2.6.5 - nok, creates '#end'
>
> Dominique Michel wrote:
> > I get the same issue yesterday.
> > This is with fvwm-2.6.6 from cvs and gentoo.
>
> So, it seems it's not a script problem.
>
> Therefore I've starting your suggestion with 'printfs' in
> instructions.c under my Fedora VM. I used this FvwmScript with one
> WriteToFile:
>
> #WindowTitle {Test WriteToFile}
> #WindowSize 470 415# Taille
> #
> #Init
> #Begin
> #Set $userDir = (GetOutput {echo "$FVWM_USERDIR"} 1 -1)
> #Set $configPath = $userDir{/bla}
> #
> #WriteToFile $configPath {This is a test}
> #Quit
> #End
>
> As I didn't knew what you meant with 'printfs' I used printf first
> and hoped that output went into .xsession-errors. No output appeared
> but WriteToFile worked. So I removed printf and compiled code
> suddenly worked!
>
> Same code, different behaviors...
>
> Therefore I did the same on my Debian 8 system and the compiled
> version of FvwmScript works also.
>
> Under Debian 8 and Fedora 19 I have distribution packages installed.
> All others are self compiled.
>
> Could it be a problem of code optimization? Because the FvwmScript
> executables has huge different sizes:
> package: 380K
> compiled: 1,5M
>
> The only thing that not fit in this to theory is Dominiques issue
> because his code is self compiled, too ...
My ebuild is the same than the one in gentoo. I modified it to use the
cvs and to be able some patches, but I don't apply them, so the code is
the one from the cvs.
The cflags are a little bit extensive, that was to be able to cross
compile in my tower and be sure to get a consistent result, but the
tower is dead now. The only ones that doesn't are for the processor are
-O2 -pipe and they should be sure:
CFLAGS="-march=amdfam10 -mcx16 -msahf -mpopcnt -mabm --param
l1-cache-size=32 --param l1-cache-line-size=64 --param
l2-cache-size=512 -mtune=amdfam10 -O2 -pipe"
CFLAGS_amd64="-m64"
CXXFLAGS="-march=amdfam10 -mcx16 -msahf -mpopcnt -mabm --param
l1-cache-size=32 --param l1-cache-line-size=64 --param
l2-cache-size=512 -mtune=amdfam10 -O2 -pipe"
LDFLAGS="-Wl,-O1 -Wl,--as-needed"
LDFLAGS_amd64="-m elf_x86_64"
The ebuild append -fno-strict-aliasing to the cflags.
The result with my configure USE flags is:
fvwm --version
fvwm 2.6.6 (from cvs) compiled on Oct 27 2013 at 16:48:42
with support for: ReadLine, RPlay, Stroke, XPM, PNG, SVG, Shape, XShm,
SM, Bidi text, XRender, XCursor, XFT, NLS
The size of FvwmScript executable is around 400.6k here.
Dominque
>
> - Thomas -
>
Re: FvwmScript - WriteToFile seems not working
2013/12/14 Dominique Michel :
> Le Sat, 14 Dec 2013 00:34:38 +0100,
> Thomas Funk a écrit :
>
>> Dan Espen wrote:
>> > Thomas Funk writes:
>> >
>> >> Hi!
>> >>
>> >> I'm working on a composite configurator with FvwmScript and
>> >> getting an issue with WriteToFile command. It only writes '#end'
>> >> into the file.
>> >>
>> >> First I thought I've done something wrong but I do the same as in
>> >> FvwmScript-BaseConfig. So I started a test with
>> >> FvwmScript-BaseConfig and the same happens. Used FVWM is 2.6.5.
>> >> This happens under Fedora 19, too.
>> >>
>> >> My next thought was something has changed in WriteToFile source
>> >> since creation of FvwmScript-BaseConfig - 2007-08-07. So I
>> >> compared the code from 2.5.22 and CVS but nothing has changed.
>> >>
>> >> So, what could be the problem?
>> >
>> > We'd probably have to see the script but you could always resort to
>> > adding printfs in Instructions.c.
>> >
>> > Start with determining if it gets NbArg right.
>> > Just seeing #end is an indication that it doesn't
>> > see the args after the filename.
>>
>> I have checked different VMs with different FVWM versions with
>> FvwmScript-BaseConfig:
>>
>> Debian 4.x (2007) with FVWM 2.5.22 - ok, creates full config
>> Debian 6.x (2011) with FVWM 2.6.0 - ok, creates full config
>> Debian 7.1 (2013) with FVWM 2.6.6 - ok, creates full config
>> Debian 7.1 (2013) with FVWM 2.6.5 - ok, creates full config
>> Debian 8 (testing, 2013) with FVWM 2.6.5 - nok, creates '#end'
>> Fedora 19 (2013) with FVWM 2.6.5 - nok, creates '#end'
>>
>> Dominique Michel wrote:
>> > I get the same issue yesterday.
>> > This is with fvwm-2.6.6 from cvs and gentoo.
>>
>> So, it seems it's not a script problem.
>>
>> Therefore I've starting your suggestion with 'printfs' in
>> instructions.c under my Fedora VM. I used this FvwmScript with one
>> WriteToFile:
>>
>> #WindowTitle {Test WriteToFile}
>> #WindowSize 470 415# Taille
>> #
>> #Init
>> #Begin
>> #Set $userDir = (GetOutput {echo "$FVWM_USERDIR"} 1 -1)
>> #Set $configPath = $userDir{/bla}
>> #
>> #WriteToFile $configPath {This is a test}
>> #Quit
>> #End
>>
>> As I didn't knew what you meant with 'printfs' I used printf first
>> and hoped that output went into .xsession-errors. No output appeared
>> but WriteToFile worked. So I removed printf and compiled code
>> suddenly worked!
>>
>> Same code, different behaviors...
>>
>> Therefore I did the same on my Debian 8 system and the compiled
>> version of FvwmScript works also.
>>
>> Under Debian 8 and Fedora 19 I have distribution packages installed.
>> All others are self compiled.
>>
>> Could it be a problem of code optimization? Because the FvwmScript
>> executables has huge different sizes:
>> package: 380K
>> compiled: 1,5M
>>
>> The only thing that not fit in this to theory is Dominiques issue
>> because his code is self compiled, too ...
>
> My ebuild is the same than the one in gentoo. I modified it to use the
> cvs and to be able some patches, but I don't apply them, so the code is
> the one from the cvs.
>
> The cflags are a little bit extensive, that was to be able to cross
> compile in my tower and be sure to get a consistent result, but the
> tower is dead now. The only ones that doesn't are for the processor are
> -O2 -pipe and they should be sure:
>
> CFLAGS="-march=amdfam10 -mcx16 -msahf -mpopcnt -mabm --param
> l1-cache-size=32 --param l1-cache-line-size=64 --param
> l2-cache-size=512 -mtune=amdfam10 -O2 -pipe"
> CFLAGS_amd64="-m64"
>
> CXXFLAGS="-march=amdfam10 -mcx16 -msahf -mpopcnt -mabm --param
> l1-cache-size=32 --param l1-cache-line-size=64 --param
> l2-cache-size=512 -mtune=amdfam10 -O2 -pipe"
>
> LDFLAGS="-Wl,-O1 -Wl,--as-needed"
> LDFLAGS_amd64="-m elf_x86_64"
>
> The ebuild append -fno-strict-aliasing to the cflags.
>
> The result with my configure USE flags is:
> fvwm --version
> fvwm 2.6.6 (from cvs) compiled on Oct 27 2013 at 16:48:42
> with support for: ReadLine, RPlay, Stroke, XPM, PNG, SVG, Shape, XShm,
> SM, Bidi text, XRender, XCursor, XFT, NLS
>
> The size of FvwmScript executable is around 400.6k here.
Is it possible, that you compile FVWM normally with configure & make, copy
the FvwmScript executable to /usr/lib/fvwm/2.6.6/ and test it with the script
snippet above or yours?
Please deactivate the debugging while configure:
./configure --cflags="-O2" (or --CFLAGS? I don't know cause I am at work ...)
@Dan: As '-g' is part of the default cflags this is the reason why my
executable
has 1,5M. Without it has around 450k. with '-O3' the size has increased and no
change of the behaviour - works as with '-O2'. So, it's not the optimize
parameter ...
- Thomas -
Re: FvwmScript - WriteToFile seems not working
Thomas Funk wrote: > @Dan: As '-g' is part of the default cflags this is the reason why my > executable has 1,5M. Without it has around 450k. with '-O3' the size > has increased and no change of the behavior - works as with '-O2'. > So, it's not the optimize parameter ... My first idea was that a newer libc6 is used to built the distribution package. So I checked the auto build log: https://buildd.debian.org/status/fetch.php?pkg=fvwm&arch=ia64&ver=1%3A2.6.5.ds-3&stamp=1384445601 And yes, they used 2.17 but the Changelog of libc6 doesn't show any Changes for the used C commands in WriteToFile between my used 2.15 and 2.17. Fedora 19 use 2.17, too and the self build works there also. Anyway I've built FVWM the Debian way to exclude the idea of problems with newer libs. With my libs the normal self compile way works but not the Debian way. Same issue with the original distribution package :( The next idea was the patches used by Chen, so I patched my working version after make clean, compiled it and ... it worked fine again. Also I used the long configure command like Debian used. Same behavior - works. Looks quiet bad ... Something get wrong while the build Debian does but no clue what's the problem ... I will investigate some time over the weekend. Hope I'll find something. By the way, Debian use -g as FVWM pretends by default. They reduce the size with strip. That's the secret ... not an optimization with -O - Thomas -
Re: FvwmScript - WriteToFile seems not working
On 19 December 2013 21:34, Thomas Funk wrote: > Also I used the long configure command like Debian used. Same behavior > - works. > > Looks quiet bad ... Something get wrong while the build Debian does > but no clue what's the problem ... Don't give out red-herrings like this---CFLAGS aside, I don't think messing with those is a good enough indication that you can just say a problem is "fixed". If you mess with CFLAGS and hence turn off/turn on higher/lower optimisations, or start dicking with -g -gdb, etc., then who knows what you'll get, it might be fine, it might not. If you want to look at this again in the future, just use "-g -O2" Having looked at the code, one optimisation that might be tweaked behind the scenes with all these CFLAGS wrangling is the fact that WriteToFile() doesn't seem to call fflush() after fwrite(). That might be OK in most cases, it depends how much buffering the kernel is doing. I wonder if you're hitting that limit and hence seeing a tiny part of the buffer fwrite() managed to get its hands on? -- Thomas Adam
