[RFC] Trying to fix testsuite/gdb.arch/i386-sse.exp pattern problem

[RFC] Trying to fix testsuite/gdb.arch/i386-sse.exp pattern problem

[Pierre Muller]

  On Cygwin, the i386-sse test always gives me 8 failures
on the int8 array values.
  Are these failure also visible on i386 linux machines?
I suppose so, but couldn't test it.

  It took me a long time to understand the complicated
regular expression used, but in the end, I suspect that
this regexpr is not adequate.

  The current test is:
     gdb_test "print \$xmm$r.v16_int8" \
        ".. = \\{(-?\[0-9\]+, ){15}-?\[0-9\]+\\}.*" \
        "check int8 contents of %xmm$r"

  But I suspect that the {15} refers to the expanded pattern,
which would mean that it would match a
 {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}
while the patterns are different (some elements are non-zero).

  I could not check this directly as
gdb return {0 <repeats 16 times>} if you set all 4 v4_float elements to
zero.

  I tried to check v8_int16 instead, but still got no luck...
Using -v 5 times, I found out that gdb_test added a group start '('
character
before the pattern and a group end ')' after the pattern, 
but even adding closing and opening groups still would not
make the test pass...

    gdb_test "set var \$xmm0.v4_float\[0\] = 0" "" "set %xmm0"
    gdb_test "set var \$xmm0.v4_float\[1\] = 0" "" "set %xmm0"
    gdb_test "set var \$xmm0.v4_float\[2\] = 0" "" "set %xmm0"
    gdb_test "set var \$xmm0.v4_float\[3\] = 0" "" "set %xmm0"
    gdb_test "print \$xmm0.v8_int16" \
        ".* =) \\{(-?\[0-9\]+, ){7}(-?\[0-9\]+\\}.*" \
        "check zeroed int8 contents of %xmm0"
Still fails, even though the result is {0, 0, 0, 0, 0, 0, 0, 0}.

Even this simple test fails:

    gdb_test "print {0, 0, 0, 0}" ".* =) \\{(0, ){3}(0\\}.*" "test array"

 I am really wondering if this pattern repetition 
is working at all...

Installed tcl is version 8.4

  I could correct the test, by putting explicitly the 
16 patterns, which results in an horribly lengthy pattern,
but I was unable to break it into pieces to have shorter source lines.

  Any ideas on ways to split the pattern string sent to gdb_test  
over multiple lines are most welcomed.


Pierre Muller

ChangeLog entry:

2007-11-07  Pierre Muller  <muller@ics.u-strasbg.fr>

	* gdb.arch/i386-see.exp: Correct pattern to check v16_int8 array of
SSE registers.


Index: i386-sse.exp
===================================================================
RCS file: /cvs/src/src/gdb/testsuite/gdb.arch/i386-sse.exp,v
retrieving revision 1.8
diff -u -p -r1.8 i386-sse.exp
--- i386-sse.exp        5 Sep 2007 00:51:49 -0000       1.8
+++ i386-sse.exp        7 Nov 2007 13:48:37 -0000
@@ -84,7 +84,7 @@ foreach r {0 1 2 3 4 5 6 7} {
         ".. = \\{$r, $r.25, $r.5, $r.75\\}.*" \
         "check float contents of %xmm$r"
     gdb_test "print \$xmm$r.v16_int8" \
-        ".. = \\{(-?\[0-9\]+, ){15}-?\[0-9\]+\\}.*" \
+        ".. = \\{-?\[0-9\]+, -?\[0-9\]+, -?\[0-9\]+, -?\[0-9\]+,
-?\[0-9\]+, -?\[0-9\]+, -?\[0-9\]+, -?\[0-9\]+, -?\[0-9\]+,-?\[0-9\]+,
-?\[0-9\]+, -?\[0-9\]+, -?\[0-9\]+, -?\[0-9\]+, -?\[0-9\]+, -?\[0-9\]+\\}.*"
\
         "check int8 contents of %xmm$r"
 }

Re: [RFC] Trying to fix testsuite/gdb.arch/i386-sse.exp pattern problem

[Daniel Jacobowitz]

On Wed, Nov 07, 2007 at 03:37:45PM +0100, Pierre Muller wrote:
> Even this simple test fails:
> 
>     gdb_test "print {0, 0, 0, 0}" ".* =) \\{(0, ){3}(0\\}.*" "test array"
> 
>  I am really wondering if this pattern repetition 
> is working at all...

Perhaps TCL relies on the system regular expression library, and
Cygwin's does not support repetition operators?  Anyway, the TCL
manual says they are supported, so I'm surprised by this.  Maybe Chris
Faylor knows something about it.

>   I could correct the test, by putting explicitly the 
> 16 patterns, which results in an horribly lengthy pattern,
> but I was unable to break it into pieces to have shorter source lines.
> 
>   Any ideas on ways to split the pattern string sent to gdb_test  
> over multiple lines are most welcomed.

I'd use variables.

set item "-?\[0-9\]+, "
set items "$item$item$item$item$item$item$item$item"
append items "$item$item$item$item$item$item$item-?\[0-9\]+"

and that's 16.

-- 
Daniel Jacobowitz
CodeSourcery

RE: [RFC] Trying to fix testsuite/gdb.arch/i386-sse.exp pattern problem

[Pierre Muller]

> -----Original Message-----
> From: gdb-patches-owner@sourceware.org [mailto:gdb-patches-
> owner@sourceware.org] On Behalf Of Daniel Jacobowitz
> Sent: Wednesday, November 07, 2007 4:11 PM
> To: Pierre Muller
> Cc: gdb-patches@sourceware.org
> Subject: Re: [RFC] Trying to fix testsuite/gdb.arch/i386-sse.exp
> pattern problem
> 
> On Wed, Nov 07, 2007 at 03:37:45PM +0100, Pierre Muller wrote:
> > Even this simple test fails:
> >
> >     gdb_test "print {0, 0, 0, 0}" ".* =) \\{(0, ){3}(0\\}.*" "test
> array"
> >
> >  I am really wondering if this pattern repetition
> > is working at all...
> 
> Perhaps TCL relies on the system regular expression library, and
> Cygwin's does not support repetition operators?  Anyway, the TCL
> manual says they are supported, so I'm surprised by this.  Maybe Chris
> Faylor knows something about it.

   I doubt this as:
if I run this:
  grep -E "[ni]{4}" gdb.log
on Cygwin, I find all lines containing the word "running"...

  So the Cygwin system regular expression seems to support
repetition operators in a way that would also match
the list of int8 with different values.

  Do I understand correctly that this is really specific
to Cygwin? 
  
> >   I could correct the test, by putting explicitly the
> > 16 patterns, which results in an horribly lengthy pattern,
> > but I was unable to break it into pieces to have shorter source
> lines.
> >
> >   Any ideas on ways to split the pattern string sent to gdb_test
> > over multiple lines are most welcomed.
> 
> I'd use variables.
> 
> set item "-?\[0-9\]+, "
> set items "$item$item$item$item$item$item$item$item"
> append items "$item$item$item$item$item$item$item-?\[0-9\]+"

  That's indeed much better already.
But if this is really a bug in the Cygwin version of
tcl/dejagnu/runtest, then it should be solved
on that end.
  Would a patch removing the wrongly interpreted
pattern (replacing it by something like Daniel just suggested)
still be acceptable?

  
Pierre

Re: [RFC] Trying to fix testsuite/gdb.arch/i386-sse.exp pattern problem

[Andreas Schwab]

"Pierre Muller" <muller@ics.u-strasbg.fr> writes:

>    I doubt this as:
> if I run this:
>   grep -E "[ni]{4}" gdb.log
> on Cygwin, I find all lines containing the word "running"...

What do you get from this?

$ echo 'send_user "[regexp {[ni]{4}} {running}]\n"' | expect

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux Products GmbH, Maxfeldstraße 5, 90409 Nürnberg, Germany
PGP key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

RE: [RFC] Trying to fix testsuite/gdb.arch/i386-sse.exp pattern problem

[Pierre Muller]

echo 'send_user "[regexp {[ni]{4}} {running}]\n"' | expect
as well as
echo 'send_user "[regexp "\[ni\]{4}" "running"]\n"' | expect

return 1...

  So expect seems to work OK...

Pierre
> -----Original Message-----
> From: gdb-patches-owner@sourceware.org [mailto:gdb-patches-
> owner@sourceware.org] On Behalf Of Andreas Schwab
> Sent: Thursday, November 08, 2007 11:44 AM
> To: Pierre Muller
> Cc: 'Daniel Jacobowitz'; gdb-patches@sourceware.org
> Subject: Re: [RFC] Trying to fix testsuite/gdb.arch/i386-sse.exp
> pattern problem
> 
> "Pierre Muller" <muller@ics.u-strasbg.fr> writes:
> 
> >    I doubt this as:
> > if I run this:
> >   grep -E "[ni]{4}" gdb.log
> > on Cygwin, I find all lines containing the word "running"...
> 
> What do you get from this?
> 
> $ echo 'send_user "[regexp {[ni]{4}} {running}]\n"' | expect
> 
> Andreas.
> 
> --
> Andreas Schwab, SuSE Labs, schwab@suse.de
> SuSE Linux Products GmbH, Maxfeldstraße 5, 90409 Nürnberg, Germany
> PGP key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276
> 4ED5
> "And now for something completely different."

Re: [RFC] Trying to fix testsuite/gdb.arch/i386-sse.exp pattern problem

['Daniel Jacobowitz']

On Thu, Nov 08, 2007 at 11:54:56AM +0100, Pierre Muller wrote:
> echo 'send_user "[regexp {[ni]{4}} {running}]\n"' | expect
> as well as
> echo 'send_user "[regexp "\[ni\]{4}" "running"]\n"' | expect
> 
> return 1...
> 
>   So expect seems to work OK...

In that case, we should figure out why that particular pattern does
not.  Try running it with runtest --debug and look at the dbg.log.

-- 
Daniel Jacobowitz
CodeSourcery

RE: [RFC] Trying to fix testsuite/gdb.arch/i386-sse.exp pattern problem

[Pierre Muller]

I simplified the test as much as I could:

$ cat ../../../src/gdb/testsuite/gdb.arch/test.exp
gdb_exit
gdb_start

gdb_test "print 1122233" \
        ".* = 1*2*.*" \
        "Check int constant"


gdb_test "print 1122233" \
        ".* = 1{1,}2{1,}.*" \
        "Check int constant with curly braces"

$ make check RUNTESTFLAGS="--debug gdb.arch/test.exp"
Nothing to be done for all...
rootme=`pwd`; export rootme; \
        srcdir=../../../src/gdb/testsuite ; export srcdir ; \
        EXPECT=`if [ -f ${rootme}/../../expect/expect ] ; then echo
${rootme}/..
/../expect/expect ; else echo expect ; fi` ; export EXPECT ; \
        EXEEXT=.exe ; export EXEEXT ; \
 
LD_LIBRARY_PATH=$rootme/../../expect:$rootme/../../libstdc++:$rootme/../
../tk/unix:$rootme/../../tcl/unix:$rootme/../../bfd:$rootme/../../opcodes:$L
D_LI
BRARY_PATH; \
        export LD_LIBRARY_PATH; \
        if [ -f ${rootme}/../../expect/expect ] ; then  \
          TCL_LIBRARY=${srcdir}/../../tcl/library ; \
          export TCL_LIBRARY ; fi ; \
        ` if [ -f ${srcdir}/../../dejagnu/runtest ]; then echo
${srcdir}/../../d
ejagnu/runtest; else if [ "i686-pc-cygwin" = "i686-pc-cygwin" ]; then echo
runte
st; else t='s,y,y,'; echo runtest | sed -e $t; fi; fi` --debug
gdb.arch/test.exp

WARNING: Couldn't find the global config file.
Test Run By Pierre on Thu Nov  8 14:18:02 2007
Native configuration is i686-pc-cygwin

                === gdb tests ===

Schedule of variations:
    unix

Running target unix
Using /usr/share/dejagnu/baseboards/unix.exp as board description file for
targe
t.
Using /usr/share/dejagnu/config/unix.exp as generic interface file for
target.
Using ../../../src/gdb/testsuite/config/unix.exp as tool-and-target-specific
int
erface file.
Running ../../../src/gdb/testsuite/gdb.arch/test.exp ...
FAIL: gdb.arch/test.exp: Check int constant with curly braces (gdb prompt
found
alone)
>>> Note that I modified gdb.log to get the (gdb prompt found alone) suffix.

                === gdb Summary ===

# of expected passes            1
# of unexpected failures        1
/usr/local/src/cvs/build/gdb/testsuite/../../gdb/gdb version
6.7.50.20071107-cv
s -nx

make: *** [just-check] Error 1

The dbg.log shows that:
expect: does "print 1122233\r\n$1 = 1122233\r\n(gdb) " (spawn_id 7) match
regula
r expression ".*A problem internal to GDB has been detected"? no^M
"\*\*\* DOSEXIT code.*"? no^M
"[\r\n]*(.* = 1*2*.*)[\r\n]+\(gdb\) $"? yes^M
But later:
expect: does "print 1122233\r\n$2 = 1122233\r\n(gdb) " (spawn_id 7) match
regula
r expression ".*A problem internal to GDB has been detected"? no^M
"\*\*\* DOSEXIT code.*"? no^M
"[\r\n]*(.* = 1{1,}2{1,}.*)[\r\n]+\(gdb\) $"? no^M

So even that simple test fails :(


Pierre

> -----Original Message-----
> From: gdb-patches-owner@sourceware.org [mailto:gdb-patches-
> owner@sourceware.org] On Behalf Of 'Daniel Jacobowitz'
> Sent: Thursday, November 08, 2007 1:38 PM
> To: Pierre Muller
> Cc: 'Andreas Schwab'; gdb-patches@sourceware.org
> Subject: Re: [RFC] Trying to fix testsuite/gdb.arch/i386-sse.exp
> pattern problem
> 
> On Thu, Nov 08, 2007 at 11:54:56AM +0100, Pierre Muller wrote:
> > echo 'send_user "[regexp {[ni]{4}} {running}]\n"' | expect
> > as well as
> > echo 'send_user "[regexp "\[ni\]{4}" "running"]\n"' | expect
> >
> > return 1...
> >
> >   So expect seems to work OK...
> 
> In that case, we should figure out why that particular pattern does
> not.  Try running it with runtest --debug and look at the dbg.log.
> 
> --
> Daniel Jacobowitz
> CodeSourcery

Re: [RFC] Trying to fix testsuite/gdb.arch/i386-sse.exp pattern problem

['Daniel Jacobowitz']

On Thu, Nov 08, 2007 at 02:30:50PM +0100, Pierre Muller wrote:
> "[\r\n]*(.* = 1{1,}2{1,}.*)[\r\n]+\(gdb\) $"? no^M
> 
> So even that simple test fails :(

How about:

echo 'send_user "[regexp {= 1{1,}2{1,}.*} {2 = 1122233}]\n"' | expect

Are you sure it's the same "expect" the Makefile is running?

-- 
Daniel Jacobowitz
CodeSourcery

RE: [RFC] Trying to fix testsuite/gdb.arch/i386-sse.exp pattern problem

[Pierre Muller]

> -----Original Message-----
> From: 'Daniel Jacobowitz' [mailto:drow@false.org]
> Sent: Thursday, November 08, 2007 2:43 PM
> To: Pierre Muller
> Cc: 'Andreas Schwab'; gdb-patches@sourceware.org
> Subject: Re: [RFC] Trying to fix testsuite/gdb.arch/i386-sse.exp
> pattern problem
> 
> On Thu, Nov 08, 2007 at 02:30:50PM +0100, Pierre Muller wrote:
> > "[\r\n]*(.* = 1{1,}2{1,}.*)[\r\n]+\(gdb\) $"? no^M
> >
> > So even that simple test fails :(
> 
> How about:
> 
> echo 'send_user "[regexp {= 1{1,}2{1,}.*} {2 = 1122233}]\n"' | expect
It also works, 
> Are you sure it's the same "expect" the Makefile is running?

Putting the send_user above inside
test.exp also gives a pattern matching...
I am lost.

Using gdb_test_multiple, I can see that the 
start and end added to the pattern of gdb_test 
is not the cause of the problem:

gdb_test_multiple  "print 1122233\n"  \
        "Check int constant with gdb_test_multiple"  \
        {  \
        -re ".*1*2*.*"  {  \
          pass "pattern found"  \
          } \
        -re ".*"  {  \
          fail "pattern not found"  \
          } \
        }

gdb_test_multiple  "print 1122233\n"  \
        "Check int constant with curly braces using gdb_test_multiple"  \
        {  \
        -re ".*1{1,}2{1,}.*"  {  \
          pass "pattern found"  \
          } \
        -re ".*"  {  \
          fail "pattern not found"  \
          } \
        }

Gives a PASS for the first and a FAIL for the second.

Pierre

Re: [RFC] Trying to fix testsuite/gdb.arch/i386-sse.exp pattern problem

['Daniel Jacobowitz']

On Thu, Nov 08, 2007 at 03:14:15PM +0100, Pierre Muller wrote:
> Putting the send_user above inside
> test.exp also gives a pattern matching...
> I am lost.

Me too.  I don't suppose this fails?

printf 'send_user "[regexp {[\r\n]*(.* = 1{1,}2{1,}.*)[\r\n]+\(gdb\) $} {print 1122233\r\n$2 = 1122233\r\n(gdb) }]\n"' | expect

It should print 1; if it prints 0 for you that would be somewhere to
start.

-- 
Daniel Jacobowitz
CodeSourcery

RE: [RFC] Trying to fix testsuite/gdb.arch/i386-sse.exp pattern problem

[Pierre Muller]

> -----Original Message-----
> From: 'Daniel Jacobowitz' [mailto:drow@false.org]
> Sent: Thursday, November 08, 2007 3:27 PM
> To: Pierre Muller
> Cc: 'Andreas Schwab'; gdb-patches@sourceware.org
> Subject: Re: [RFC] Trying to fix testsuite/gdb.arch/i386-sse.exp
> pattern problem
> 
> On Thu, Nov 08, 2007 at 03:14:15PM +0100, Pierre Muller wrote:
> > Putting the send_user above inside
> > test.exp also gives a pattern matching...
> > I am lost.
> 
> Me too.  I don't suppose this fails?
> 
> printf 'send_user "[regexp {[\r\n]*(.* = 1{1,}2{1,}.*)[\r\n]+\(gdb\) $}
> {print 1122233\r\n$2 = 1122233\r\n(gdb) }]\n"' | expect
It does succeed indeed... 
> It should print 1; if it prints 0 for you that would be somewhere to
> start.

  But if I put the send_user part inside my test.exp
it returns 0!

Pierre

Re: [RFC] Trying to fix testsuite/gdb.arch/i386-sse.exp pattern problem

['Daniel Jacobowitz']

On Thu, Nov 08, 2007 at 03:51:11PM +0100, Pierre Muller wrote:
> > printf 'send_user "[regexp {[\r\n]*(.* = 1{1,}2{1,}.*)[\r\n]+\(gdb\) $}
> > {print 1122233\r\n$2 = 1122233\r\n(gdb) }]\n"' | expect
> It does succeed indeed... 
> > It should print 1; if it prints 0 for you that would be somewhere to
> > start.
> 
>   But if I put the send_user part inside my test.exp
> it returns 0!

That's not surprising.  The quoting rules are different.  Try:

send_user "[regexp {[\r\n]*(.* = 1{1,}2{1,}.*)[\r\n]+\(gdb\) $} \
	  "print 1122233\r\n\$2 = 1122233\r\n(gdb) "]\n"

The quotes instead of braces let literal \r\n get into the test string.

-- 
Daniel Jacobowitz
CodeSourcery