Re: [PATCH] quoting curly braces in call-rt-st

Re: [PATCH] quoting curly braces in call-rt-st

[Michael Elizabeth Chastain]

Here is how I build my Dejagnu, as of this weekend's test run.

First I have baseline versions of TCL and Expect:

  ftp://tcl.activestate.com/pub/tcl/tcl8_0/tcl8.0.5.tar.gz
  http://expect.nist.gov/src/expect-5.30.tar.gz

These are not the latest versions of TCL and Expect, but they
work for me.

I configured TCL with "--enable-gcc" and "--disable-shared".
I configured Expect with "--disable-shared", "--with-tcl",
  "--with-tclconfig", and "--with-tclinclude".

These are installed on all the machines I use so I don't have to rebuild
them every week.

When I check out gdb, I also check out naked-dejagnu:

  CVSROOT=:pserver:anoncvs@anoncvs.cygnus.com:/cvs/src
  export CVSROOT
  expect -c 'spawn cvs login' -c 'expect password: ' -c 'send -- "anoncvs\r"' -c 'expect eof'
  cvs -z 9 co gdb binutils naked-dejagnu
  mv src source-src
  rm -rf CVS

I don't configure or make dejagnu.  When I run the gdb testsuite, it
looks for dejagnu/runtest in the source tree and finds it there.

(binutils is there because I'm also building gcc from cvs source).

My inclination is to leave the curly braces alone and just use the
known good test machinery for now.

In the future I would like to switch from the "naked-dejagnu" module
to some stable version of dejagnu.  And I'd like to firm up my scripts
and make them available to other people.  And I'd like to run the
gdb test script and have 0 ERRORs, 0 WARNINGs, and less than 10 FAILs.

MichaelC

Re: [PATCH] quoting curly braces in call-rt-st

[Fernando Nasser]

Michael Snyder wrote:
> 
> Using double-backslash to quote curly braces in regular expressions
> seems so satisfy the largest subset of versions of expect.
> 

If I understand right, Michael claims his double-backslashed version works with both incarnations of expect regexps.  So, we have a perfectly valid pattern which is more robust (works independently of the expect version).

I think we should make this change anyway.

                                                  
-- 
Fernando Nasser
Red Hat - Toronto                       E-Mail:  fnasser@redhat.com
2323 Yonge Street, Suite #300
Toronto, Ontario   M4P 2C9

Re: [PATCH] quoting curly braces in call-rt-st

[H . J . Lu]

On Mon, Jul 09, 2001 at 04:37:54PM -0400, Fernando Nasser wrote:
> Michael Snyder wrote:
> > 
> > Using double-backslash to quote curly braces in regular expressions
> > seems so satisfy the largest subset of versions of expect.
> > 
> 
> If I understand right, Michael claims his double-backslashed version works with both incarnations of expect regexps.  So, we have a perfectly valid pattern which is more robust (works independently of the expect version).
> 
> I think we should make this change anyway.
> 

See

http://gcc.gnu.org/ml/gcc-patches/2001-07/msg00034.html
http://gcc.gnu.org/ml/gcc-patches/2000-09/msg00353.html


H.J.

Re: [PATCH] quoting curly braces in call-rt-st

[Fernando Nasser]

"H . J . Lu" wrote:
> 
> On Mon, Jul 09, 2001 at 04:37:54PM -0400, Fernando Nasser wrote:
> > Michael Snyder wrote:
> > >
> > > Using double-backslash to quote curly braces in regular expressions
> > > seems so satisfy the largest subset of versions of expect.
> > >
> >
> > If I understand right, Michael claims his double-backslashed version works with both incarnations of expect regexps.  So, we have a perfectly valid pattern which is more robust (works independently of the expect version).
> >
> > I think we should make this change anyway.
> >
> 
> See
> 
> http://gcc.gnu.org/ml/gcc-patches/2001-07/msg00034.html
> http://gcc.gnu.org/ml/gcc-patches/2000-09/msg00353.html
> 
> H.J.

Thank you H.J.

Folks, we may want to do the same on other test files...



-- 
Fernando Nasser
Red Hat - Toronto                       E-Mail:  fnasser@redhat.com
2323 Yonge Street, Suite #300
Toronto, Ontario   M4P 2C9

Re: [PATCH] quoting curly braces in call-rt-st

[Michael Snyder]

Fernando Nasser wrote:
> 
> "H . J . Lu" wrote:
> >
> > On Mon, Jul 09, 2001 at 04:37:54PM -0400, Fernando Nasser wrote:
> > > Michael Snyder wrote:
> > > >
> > > > Using double-backslash to quote curly braces in regular expressions
> > > > seems so satisfy the largest subset of versions of expect.
> > > >
> > >
> > > If I understand right, Michael claims his double-backslashed version works with both incarnations of expect regexps.  So, we have a perfectly valid pattern which is more robust (works independently of the expect version).
> > >
> > > I think we should make this change anyway.
> > >
> >
> > See
> >
> > http://gcc.gnu.org/ml/gcc-patches/2001-07/msg00034.html
> > http://gcc.gnu.org/ml/gcc-patches/2000-09/msg00353.html
> >
> > H.J.
> 
> Thank you H.J.
> 
> Folks, we may want to do the same on other test files...

By "do the same", do you mean "use double backslash to quote "{"/"}"?
If so, I agree with you.

Re: [PATCH] quoting curly braces in call-rt-st

[Fernando Nasser]

Michael Snyder wrote:
> 
> Fernando Nasser wrote:
> >
> > "H . J . Lu" wrote:
> > >
> > > On Mon, Jul 09, 2001 at 04:37:54PM -0400, Fernando Nasser wrote:
> > > > Michael Snyder wrote:
> > > > >
> > > > > Using double-backslash to quote curly braces in regular expressions
> > > > > seems so satisfy the largest subset of versions of expect.
> > > > >
> > > >
> > > > If I understand right, Michael claims his double-backslashed version works with both incarnations of expect regexps.  So, we have a perfectly valid pattern which is more robust (works independently of the expect version).
> > > >
> > > > I think we should make this change anyway.
> > > >
> > >
> > > See
> > >
> > > http://gcc.gnu.org/ml/gcc-patches/2001-07/msg00034.html
> > > http://gcc.gnu.org/ml/gcc-patches/2000-09/msg00353.html
> > >
> > > H.J.
> >
> > Thank you H.J.
> >
> > Folks, we may want to do the same on other test files...
> 
> By "do the same", do you mean "use double backslash to quote "{"/"}"?
> If so, I agree with you.

Yes.

-- 
Fernando Nasser
Red Hat - Toronto                       E-Mail:  fnasser@redhat.com
2323 Yonge Street, Suite #300
Toronto, Ontario   M4P 2C9

Re: [PATCH] quoting curly braces in call-rt-st

[Michael Snyder]

Fernando Nasser wrote:
> 
> Michael Snyder wrote:
> >
> > Fernando Nasser wrote:
> > >
> > > "H . J . Lu" wrote:
> > > >
> > > > On Mon, Jul 09, 2001 at 04:37:54PM -0400, Fernando Nasser wrote:
> > > > > Michael Snyder wrote:
> > > > > >
> > > > > > Using double-backslash to quote curly braces in regular expressions
> > > > > > seems so satisfy the largest subset of versions of expect.
> > > > > >
> > > > >
> > > > > If I understand right, Michael claims his double-backslashed version works with both incarnations of expect regexps.  So, we have a perfectly valid pattern which is more robust (works independently of the expect version).
> > > > >
> > > > > I think we should make this change anyway.
> > > > >
> > > >
> > > > See
> > > >
> > > > http://gcc.gnu.org/ml/gcc-patches/2001-07/msg00034.html
> > > > http://gcc.gnu.org/ml/gcc-patches/2000-09/msg00353.html
> > > >
> > > > H.J.
> > >
> > > Thank you H.J.
> > >
> > > Folks, we may want to do the same on other test files...
> >
> > By "do the same", do you mean "use double backslash to quote "{"/"}"?
> > If so, I agree with you.
> 
> Yes.

COmmitted.

Re: [PATCH] quoting curly braces in call-rt-st

[Michael Elizabeth Chastain]

I think the patterns are breaking now because a different person is
running them in a different environment -- specifically, with a different
version of 'expect'.  I always use /usr/progressive/bin/expect
on every platform.  I'm not sure what MichaelS is using; it depends
on his path.

This means our test harness machinery is vulnerable and fragile.
That's not good.  :(

I did some grepping:

  % cd /horton/chastain/fsf/log
  % grep -r 'FAIL.*call-rt-st.exp' *

My only FAIL's or XFAIL's are due to that new "procfs" message.

> For what its worth I was having fun running expect on Red Hat 7.x. 
> Zapping the expect built from CVS so that the pre-installed expect was 
> used mysteriously eliminated the problem.

That's weird because I get *more* fails when I use /usr/bin/expect!

Do you have /usr/progressive/bin in your path in front of /usr/bin?

BTW, I changed my test script recently so that it creates a new directory,
explicitly copies in the version of "expect" that I want to use,
and sticks that directory at the head of $PATH.

> PS: As an aside, I know Ian Rox. is thinking of hardwiring expects 
> pattern matcher because the tcl8.0->tcl8.3 transition results in broken 
> patterns.

I saw that message going by.  I think that would help.  Tom Tromey also
says that our gdb_test stack is nasty code.

MichaelC

Re: [PATCH] quoting curly braces in call-rt-st

[Andrew Cagney]

> I think the patterns are breaking now because a different person is
> running them in a different environment -- specifically, with a different
> version of 'expect'.  I always use /usr/progressive/bin/expect
> on every platform.  I'm not sure what MichaelS is using; it depends
> on his path.


(Brain gets out of first gear...)  Yes, if I zap the expect built from 
the CVS repository I'm going to get the expect in progressive

(For those wondering what progressive is, it is a Red Hat standardized 
tested/built version of GCC+BINUTILS+GDB+...)


> This means our test harness machinery is vulnerable and fragile.
> That's not good.  :(


I think it means that expect has a poorly defined regex syntax and now 
it is changing under our feet.  Once a an expression is defined it 
should just keep working.

	Andrew

Re: [PATCH] quoting curly braces in call-rt-st

[Michael Elizabeth Chastain]

I proofread this.  One problem: it needs "2001" in the list of
copyright years on the first line.  Other than that, it looked correct
and complete.  I checked for any residual "\{" or "\}"; there were none.

I tested this on native i686-pc-linux-gnu (Red Hat Linux 7) and
native sparc-sun-solaris2.8.  On Red Hat Linux 7, I used both
/usr/bin/expect and Gnupro expect.  It worked in all setups.

Recommended for approval, with the copyright year added to it.

MichaelC

Re: [PATCH] quoting curly braces in call-rt-st

[Andrew Cagney]

Um,

Why are these patterns mysteriously breaking now.  call-ar-st has been 
in the repository for a very long time.

For what its worth I was having fun running expect on Red Hat 7.x. 
Zapping the expect built from CVS so that the pre-installed expect was 
used mysteriously eliminated the problem.

If this is trying to address that same problem then I don't think it is 
a bug in GDB's patterns.  Rather it is a bug in how expect was being built.

	Andrew

PS: As an aside, I know Ian Rox. is thinking of hardwiring expects 
pattern matcher because the tcl8.0->tcl8.3 transition results in broken 
patterns.

Re: [PATCH] quoting curly braces in call-rt-st

[Michael Snyder]

Andrew Cagney wrote:
> 
> Um,
> 
> Why are these patterns mysteriously breaking now.  call-ar-st has been
> in the repository for a very long time.

I don't think this is anything new.  Call-ar-st has always been
a source of mysterious and erratic fails (at least for me).

> 
> For what its worth I was having fun running expect on Red Hat 7.x.
> Zapping the expect built from CVS so that the pre-installed expect was
> used mysteriously eliminated the problem.

I know it's an expect issue -- I'm just trying to work around it.

> If this is trying to address that same problem then I don't think it is
> a bug in GDB's patterns.  Rather it is a bug in how expect was being built.

Yep, it's a work-around for an expect issue.  But we do that
all the time.