Help: Make going wild

Help: Make going wild

[Tadeusz Liszka]

I have make calling user shell script, calling VC++ compiled code,
calling exec, calling rm and then (sometimes, but too often) rm goes
wild (takes all cpu time, grows in size to several megabytes) and I have
to kill it. Sometimes it is cp not rm.

Any suggestions ? I suspect CYGWIN with NT4sp4 because it happens on all
machines with sp4 but not on those with sp3.

Details:

I have big Unix application, FORTRAN, C, about 20MB of source, Tcl/Tk,
OpenGL, lex/yacc, and our own preprocessing on top of it. When porting
to NT I decided on redoing completely the make structure (old one was
way too complex) and, after several abortive attempts (long and bloody
Microsoft bashing skipped) zoomed on CYGWIN32, using gnu-make and gcc
for preprocessing, but using VC++ and PowerFortran for actual
compilation (do not ask who made this decision :). CYGWIN works GREAT.
The differences between the platforms are minimal (mainly compiler
options, renaming .obj .lib .exe, etc). Sometimes at the end of make I
got error message (sigbus or such), but it never affected the works.

This month I took this code and all make structure to another office and
installed it on their machines, HP Kayak, NT4, the main difference in
setup is that they upgraded to service pack 4. On their machines it
takes 4-5 attempts on the average to go through preprocessing phase (our
own code, generated by lex/yacc, and using lotsa exec's to move files
around). This code and the script calling it is fairly old (== tested
and trusted), I used MKS lex/yacc/VC5++ to build it, just because I need
it anyway in other place in make - I guess I could try to build it using
CYGWIN (bison, gcc) but it does not seem to be the culprit.

I have installed full.exe with all default settings, no other setup
except MAKE_MODE=UNIX and PATH, default mounts (text!=binary). The
problem happens when I execute make from any: DOS/tcsh/bash, there is
plenty of RAM and disk space.

How can I check if my cygwin is up-to-date ? Has anybody seen such
problems ? Are there some patches already ?

I subscribe to to the list, and was looking for related topics, but
found none.
-- 
Tadeusz
:: The public opinion should be alarmed by its own nonexistence
:: (512)467-0618 ext. 526 ::       Stanislaw J. Lec, trans. TJL

Re: Help: Make going wild

[Tadeusz Liszka]

Few more data items:

After installation of newest full.exe the problem is 100% persistent,
but it does not happen on the first exec in the sequence

Following is from the good, working setup (NT4sp3) and it may not be
pertinent to the problem

The intermittent error message I was getting sometimes is:

proc: c:\cygnus....\make.EXE 1000 (0) sig_send: wait for sig_complete
event failed, sig 20, rc -1, error 6

when I try to break during that preprocessing I get in adddition
following message

[main] make 1000 (0) proc_terminated WFSO (hwait_subproc), rc -1, error
288

Besides messages nothing suggested that anything went wrong

Tadeusz Liszka wrote:
> 
> I have make calling user shell script, calling VC++ compiled code,
> calling exec, calling rm and then (sometimes, but too often) rm goes
> wild (takes all cpu time, grows in size to several megabytes) and I have
> to kill it. Sometimes it is cp not rm.
> 
> Any suggestions ? I suspect CYGWIN with NT4sp4 because it happens on all
> machines with sp4 but not on those with sp3.
> 
> Details:
> 
> I have big Unix application, FORTRAN, C, about 20MB of source, Tcl/Tk,
> OpenGL, lex/yacc, and our own preprocessing on top of it. When porting
> to NT I decided on redoing completely the make structure (old one was
> way too complex) and, after several abortive attempts (long and bloody
> Microsoft bashing skipped) zoomed on CYGWIN32, using gnu-make and gcc
> for preprocessing, but using VC++ and PowerFortran for actual
> compilation (do not ask who made this decision :). CYGWIN works GREAT.
> The differences between the platforms are minimal (mainly compiler
> options, renaming .obj .lib .exe, etc). Sometimes at the end of make I
> got error message (sigbus or such), but it never affected the works.
> 
> This month I took this code and all make structure to another office and
> installed it on their machines, HP Kayak, NT4, the main difference in
> setup is that they upgraded to service pack 4. On their machines it
> takes 4-5 attempts on the average to go through preprocessing phase (our
> own code, generated by lex/yacc, and using lotsa exec's to move files
> around). This code and the script calling it is fairly old (== tested
> and trusted), I used MKS lex/yacc/VC5++ to build it, just because I need
> it anyway in other place in make - I guess I could try to build it using
> CYGWIN (bison, gcc) but it does not seem to be the culprit.
> 
> I have installed full.exe with all default settings, no other setup
> except MAKE_MODE=UNIX and PATH, default mounts (text!=binary). The
> problem happens when I execute make from any: DOS/tcsh/bash, there is
> plenty of RAM and disk space.
> 
> How can I check if my cygwin is up-to-date ? Has anybody seen such
> problems ? Are there some patches already ?
> 
> I subscribe to to the list, and was looking for related topics, but
> found none.
> --
> Tadeusz
> :: The public opinion should be alarmed by its own nonexistence
> :: (512)467-0618 ext. 526 ::       Stanislaw J. Lec, trans. TJL

-- 
Tadeusz
:: The public opinion should be alarmed by its own nonexistence
:: (512)467-0618 ext. 526 ::       Stanislaw J. Lec, trans. TJL

Re: Help: Make going wild

[Geoffrey Noer]

On Tue, Feb 02, 1999, Tadeusz Liszka wrote:
[...]
> proc: c:\cygnus....\make.EXE 1000 (0) sig_send: wait for sig_complete
> event failed, sig 20, rc -1, error 6
> 
> when I try to break during that preprocessing I get in adddition
> following message
> 
> [main] make 1000 (0) proc_terminated WFSO (hwait_subproc), rc -1, error
> 288

Are you running Cygwin 20.1?  My NT machine has been running NTsp4 for
quite some time; I haven't noticed any change of behavior with Cygwin.

-- 
Geoffrey Noer
noer@cygnus.com

Re: Help: Make going wild

[Greg Miller]

Tadeusz Liszka wrote:
> Following is from the good, working setup (NT4sp3) and it may not be
> pertinent to the problem
> 
> The intermittent error message I was getting sometimes is:
> 
> proc: c:\cygnus....\make.EXE 1000 (0) sig_send: wait for sig_complete
> event failed, sig 20, rc -1, error 6

I get this all the time on my setup with NT4sp4, but it occurs after the
final executable is produced and appears to be harmless. I suspect it's
unrelated to your main problem.
-- 
http://www.classic-games.com/
If Bill Clinton doesn't have to obey the law, neither do you.
*** NEWBIES: Limit signatures to four lines! No HTML mail or posts! ***

Re: Help: Make going wild

[Tadeusz Liszka]

Thank you very much for the responses to my mail.

I have got two confirmations that the cygwin+NT4sp4 is cool, but
unfortunately I am positive that this is the source of the problem in my
case. Apparently Microsoft investigated CYGNUS and managed to slip some
goodie into sp4 (:-)

So the situation again:

make calls the sh script, which calls dsmake, which uses several exec's
to manipulate files (cat, cp, rm)

Dsmake is our proprietary preprocessor, built using lex/yacc

Initially I compiled dsmake using VC++ (and MKS lex/yacc). This worked
ok in NT4sp3.

When I switched to sp4 (reproduced on at least three different machines)
some of the exec'd Unix functions go wild (take 99%CPU, grow in size
~50K/sec, and never quit). It is random: sometimes it is rm sometimes cp
I have even seen cmd.exe doing it !. Everytime this happens there is
empty line printed to the screen (?)

After recompilation of dsmake using CYGNUS tools (lex/yacc/gcc)
everything so far works as expected.

I see following possible explanations:
1. sp4 changed the exec system call in a way incompatible with CYGWIN
2. dsmake compiled with sp3 will not work with sp4
3. conceivably there could be error in VC++ or MKS lex/yacc, which shows
only after seemingly unrelated sp4 update
4. I ruled out error in dsmake itself (it's been around for a while on
several Unixes, the logic is simple, and has been debugged and purified
carefully), but there may be some lex/yacc non-portable stuff, which
shows on MKS lex/yacc only

I have no resources nor enough CYGNUS experience to investigate error,
if the new recompiled dsmake passes all tests - I will drop the issue.
However option No 1 above is my best bet - as such it may warrant
investigations by somebody. I can reproduce problem and send
nonproprietary part of the code if there is interest.

regards


Greg Miller wrote:
> 
> Tadeusz Liszka wrote:
> > Following is from the good, working setup (NT4sp3) and it may not be
> > pertinent to the problem
> >
> > The intermittent error message I was getting sometimes is:
> >
> > proc: c:\cygnus....\make.EXE 1000 (0) sig_send: wait for sig_complete
> > event failed, sig 20, rc -1, error 6
> 
> I get this all the time on my setup with NT4sp4, but it occurs after the
> final executable is produced and appears to be harmless. I suspect it's
> unrelated to your main problem.
> --
> http://www.classic-games.com/
> If Bill Clinton doesn't have to obey the law, neither do you.
> *** NEWBIES: Limit signatures to four lines! No HTML mail or posts! ***

-- 
Tadeusz
:: The public opinion should be alarmed by its own nonexistence
:: (512)467-0618 ext. 526 ::       Stanislaw J. Lec, trans. TJL

Re: Help: Make going wild

[Tadeusz Liszka]

Thank you very much for the responses to my mail.

I have got two confirmations that the cygwin+NT4sp4 is cool, but
unfortunately I am positive that this is the source of the problem in my
case. Apparently Microsoft investigated CYGNUS and managed to slip some
goodie into sp4 (:-)

So the situation again:

make calls the sh script, which calls dsmake, which uses several exec's
to manipulate files (cat, cp, rm)

Dsmake is our proprietary preprocessor, built using lex/yacc

Initially I compiled dsmake using VC++ (and MKS lex/yacc). This worked
ok in NT4sp3.

When I switched to sp4 (reproduced on at least three different machines)
some of the exec'd Unix functions go wild (take 99%CPU, grow in size
~50K/sec, and never quit). It is random: sometimes it is rm sometimes cp
I have even seen cmd.exe doing it !. Everytime this happens there is
empty line printed to the screen (?)

After recompilation of dsmake using CYGNUS tools (lex/yacc/gcc)
everything so far works as expected.

I see following possible explanations:
1. sp4 changed the exec system call in a way incompatible with CYGWIN
2. dsmake compiled with sp3 will not work with sp4
3. conceivably there could be error in VC++ or MKS lex/yacc, which shows
only after seemingly unrelated sp4 update
4. I ruled out error in dsmake itself (it's been around for a while on
several Unixes, the logic is simple, and has been debugged and purified
carefully), but there may be some lex/yacc non-portable stuff, which
shows on MKS lex/yacc only

I have no resources nor enough CYGNUS experience to investigate error,
if the new recompiled dsmake passes all tests - I will drop the issue.
However option No 1 above is my best bet - as such it may warrant
investigations by somebody. I can reproduce problem and send
nonproprietary part of the code if there is interest.

regards


Greg Miller wrote:
> 
> Tadeusz Liszka wrote:
> > Following is from the good, working setup (NT4sp3) and it may not be
> > pertinent to the problem
> >
> > The intermittent error message I was getting sometimes is:
> >
> > proc: c:\cygnus....\make.EXE 1000 (0) sig_send: wait for sig_complete
> > event failed, sig 20, rc -1, error 6
> 
> I get this all the time on my setup with NT4sp4, but it occurs after the
> final executable is produced and appears to be harmless. I suspect it's
> unrelated to your main problem.
> --
> http://www.classic-games.com/
> If Bill Clinton doesn't have to obey the law, neither do you.
> *** NEWBIES: Limit signatures to four lines! No HTML mail or posts! ***

-- 
Tadeusz
:: The public opinion should be alarmed by its own nonexistence
:: (512)467-0618 ext. 526 ::       Stanislaw J. Lec, trans. TJL

Re: Help: Make going wild

[Geoffrey Noer]

On Tue, Feb 02, 1999, Tadeusz Liszka wrote:
[...]
> proc: c:\cygnus....\make.EXE 1000 (0) sig_send: wait for sig_complete
> event failed, sig 20, rc -1, error 6
> 
> when I try to break during that preprocessing I get in adddition
> following message
> 
> [main] make 1000 (0) proc_terminated WFSO (hwait_subproc), rc -1, error
> 288

Are you running Cygwin 20.1?  My NT machine has been running NTsp4 for
quite some time; I haven't noticed any change of behavior with Cygwin.

-- 
Geoffrey Noer
noer@cygnus.com

Re: Help: Make going wild

[Greg Miller]

Tadeusz Liszka wrote:
> Following is from the good, working setup (NT4sp3) and it may not be
> pertinent to the problem
> 
> The intermittent error message I was getting sometimes is:
> 
> proc: c:\cygnus....\make.EXE 1000 (0) sig_send: wait for sig_complete
> event failed, sig 20, rc -1, error 6

I get this all the time on my setup with NT4sp4, but it occurs after the
final executable is produced and appears to be harmless. I suspect it's
unrelated to your main problem.
-- 
http://www.classic-games.com/
If Bill Clinton doesn't have to obey the law, neither do you.
*** NEWBIES: Limit signatures to four lines! No HTML mail or posts! ***

Help: Make going wild

[Tadeusz Liszka]

I have make calling user shell script, calling VC++ compiled code,
calling exec, calling rm and then (sometimes, but too often) rm goes
wild (takes all cpu time, grows in size to several megabytes) and I have
to kill it. Sometimes it is cp not rm.

Any suggestions ? I suspect CYGWIN with NT4sp4 because it happens on all
machines with sp4 but not on those with sp3.

Details:

I have big Unix application, FORTRAN, C, about 20MB of source, Tcl/Tk,
OpenGL, lex/yacc, and our own preprocessing on top of it. When porting
to NT I decided on redoing completely the make structure (old one was
way too complex) and, after several abortive attempts (long and bloody
Microsoft bashing skipped) zoomed on CYGWIN32, using gnu-make and gcc
for preprocessing, but using VC++ and PowerFortran for actual
compilation (do not ask who made this decision :). CYGWIN works GREAT.
The differences between the platforms are minimal (mainly compiler
options, renaming .obj .lib .exe, etc). Sometimes at the end of make I
got error message (sigbus or such), but it never affected the works.

This month I took this code and all make structure to another office and
installed it on their machines, HP Kayak, NT4, the main difference in
setup is that they upgraded to service pack 4. On their machines it
takes 4-5 attempts on the average to go through preprocessing phase (our
own code, generated by lex/yacc, and using lotsa exec's to move files
around). This code and the script calling it is fairly old (== tested
and trusted), I used MKS lex/yacc/VC5++ to build it, just because I need
it anyway in other place in make - I guess I could try to build it using
CYGWIN (bison, gcc) but it does not seem to be the culprit.

I have installed full.exe with all default settings, no other setup
except MAKE_MODE=UNIX and PATH, default mounts (text!=binary). The
problem happens when I execute make from any: DOS/tcsh/bash, there is
plenty of RAM and disk space.

How can I check if my cygwin is up-to-date ? Has anybody seen such
problems ? Are there some patches already ?

I subscribe to to the list, and was looking for related topics, but
found none.
-- 
Tadeusz
:: The public opinion should be alarmed by its own nonexistence
:: (512)467-0618 ext. 526 ::       Stanislaw J. Lec, trans. TJL

Re: Help: Make going wild

[Tadeusz Liszka]

Few more data items:

After installation of newest full.exe the problem is 100% persistent,
but it does not happen on the first exec in the sequence

Following is from the good, working setup (NT4sp3) and it may not be
pertinent to the problem

The intermittent error message I was getting sometimes is:

proc: c:\cygnus....\make.EXE 1000 (0) sig_send: wait for sig_complete
event failed, sig 20, rc -1, error 6

when I try to break during that preprocessing I get in adddition
following message

[main] make 1000 (0) proc_terminated WFSO (hwait_subproc), rc -1, error
288

Besides messages nothing suggested that anything went wrong

Tadeusz Liszka wrote:
> 
> I have make calling user shell script, calling VC++ compiled code,
> calling exec, calling rm and then (sometimes, but too often) rm goes
> wild (takes all cpu time, grows in size to several megabytes) and I have
> to kill it. Sometimes it is cp not rm.
> 
> Any suggestions ? I suspect CYGWIN with NT4sp4 because it happens on all
> machines with sp4 but not on those with sp3.
> 
> Details:
> 
> I have big Unix application, FORTRAN, C, about 20MB of source, Tcl/Tk,
> OpenGL, lex/yacc, and our own preprocessing on top of it. When porting
> to NT I decided on redoing completely the make structure (old one was
> way too complex) and, after several abortive attempts (long and bloody
> Microsoft bashing skipped) zoomed on CYGWIN32, using gnu-make and gcc
> for preprocessing, but using VC++ and PowerFortran for actual
> compilation (do not ask who made this decision :). CYGWIN works GREAT.
> The differences between the platforms are minimal (mainly compiler
> options, renaming .obj .lib .exe, etc). Sometimes at the end of make I
> got error message (sigbus or such), but it never affected the works.
> 
> This month I took this code and all make structure to another office and
> installed it on their machines, HP Kayak, NT4, the main difference in
> setup is that they upgraded to service pack 4. On their machines it
> takes 4-5 attempts on the average to go through preprocessing phase (our
> own code, generated by lex/yacc, and using lotsa exec's to move files
> around). This code and the script calling it is fairly old (== tested
> and trusted), I used MKS lex/yacc/VC5++ to build it, just because I need
> it anyway in other place in make - I guess I could try to build it using
> CYGWIN (bison, gcc) but it does not seem to be the culprit.
> 
> I have installed full.exe with all default settings, no other setup
> except MAKE_MODE=UNIX and PATH, default mounts (text!=binary). The
> problem happens when I execute make from any: DOS/tcsh/bash, there is
> plenty of RAM and disk space.
> 
> How can I check if my cygwin is up-to-date ? Has anybody seen such
> problems ? Are there some patches already ?
> 
> I subscribe to to the list, and was looking for related topics, but
> found none.
> --
> Tadeusz
> :: The public opinion should be alarmed by its own nonexistence
> :: (512)467-0618 ext. 526 ::       Stanislaw J. Lec, trans. TJL

-- 
Tadeusz
:: The public opinion should be alarmed by its own nonexistence
:: (512)467-0618 ext. 526 ::       Stanislaw J. Lec, trans. TJL