[patch/rfc] post-process the `maint print architecture' from gdb_mbuild.sh

[patch/rfc] post-process the `maint print architecture' from gdb_mbuild.sh

[Andrew Cagney]

[-- Attachment #1: Type: text/plain, Size: 531 bytes --]

Hello,

The attached patch modifies the gdb_mbuild.sh so that it post-processes 
the output from each target's `maint print architecture' command.

The post processing does two things:

- (using addr2line) replaces <0xNNNN> with the name of the corresponding 
function / variable

- strips off any leading path info included in any file name paths

Doing this makes the task of comparing the `maint print architecture' 
output from separate gdb_mbuild.sh runs easier - functions and file 
names are identical.

thoughts?

Andrew



[-- Attachment #2: diffs --]
[-- Type: text/plain, Size: 1266 bytes --]

2003-01-02  Andrew Cagney  <ac131313@redhat.com>

	* gdb_mbuild.sh: Edit the output of `maint print architecture'
	replacing hex constants with function names and stripping leading
	file name directory prefixes.

Index: gdb_mbuild.sh
===================================================================
RCS file: /cvs/src/src/gdb/gdb_mbuild.sh,v
retrieving revision 1.3
diff -u -r1.3 gdb_mbuild.sh
--- gdb_mbuild.sh	2 Jan 2003 16:40:33 -0000	1.3
+++ gdb_mbuild.sh	2 Jan 2003 19:16:23 -0000
@@ -279,6 +279,31 @@
     fail "gdb printed no output" ! -s Gdb.log
     grep -e internal-error Gdb.log && fail "gdb panic" 1
 
+    # Parse the architecture dump replacing any <0xNNNN> with the
+    # corresponding function.
+
+    sed -n \
+	-e '/<0x0*>/d' \
+	-e 's/^.*<0x\([0-9a-f]*\)>.*$/0x\1/p' \
+	Gdb.log | sort -u | while read addr
+    do
+	func="`addr2line -f -e ./gdb/gdb -s ${addr} | sed -n -e 1p`"
+	echo ${addr} ${func}
+	ed -s Gdb.log <<EOF
+, s/<${addr}>/<${func}>/g
+w
+q
+EOF
+    done
+
+    # Strip out the leading string from any file names that
+    # contain a prefix.
+    ed -s Gdb.log <<EOF
+, s/"\/.*\/gdb\//"gdb\//g
+w
+q
+EOF
+
     # Replace the build directory with a file as semaphore that stops
     # a rebuild. (should the logs be saved?)
 

Re: [patch/rfc] post-process the `maint print architecture' from gdb_mbuild.sh

[Michael Elizabeth Chastain]

> Thoughts?

The idea sounds good to me.

Instead of ed'ing Gdb.log, how about just naming the first file
Gdb.out or Gdb-raw.log, and then sed'ing into Gdb.log.

Michael C

Re: [patch/rfc] post-process the `maint print architecture' from gdb_mbuild.sh

[Andrew Cagney]

>> Thoughts?
> 
> 
> The idea sounds good to me.
> 
> Instead of ed'ing Gdb.log, how about just naming the first file
> Gdb.out or Gdb-raw.log, and then sed'ing into Gdb.log.

I can save the unedited output.

To make sed worthwhile, I'd need to construct a sed script that did all 
substitutions in a single pass.  Otherwize the script ends up repeating:
	sed -e ... < input > output
	mv output input
which is equivalent to ed.  That, I think, is something for a rainy day.

Andrew

Re: [patch/rfc] post-process the `maint print architecture' from gdb_mbuild.sh

[Daniel Jacobowitz]

On Tue, Jan 07, 2003 at 10:10:36AM -0500, Andrew Cagney wrote:
> >>Thoughts?
> >
> >
> >The idea sounds good to me.
> >
> >Instead of ed'ing Gdb.log, how about just naming the first file
> >Gdb.out or Gdb-raw.log, and then sed'ing into Gdb.log.
> 
> I can save the unedited output.
> 
> To make sed worthwhile, I'd need to construct a sed script that did all 
> substitutions in a single pass.  Otherwize the script ends up repeating:
> 	sed -e ... < input > output
> 	mv output input
> which is equivalent to ed.  That, I think, is something for a rainy day.

It's as simple as "sed -e ... -e ... -e ... -e ... < input > output",
isn't it?  Or via -f and a file, the same thing.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: [patch/rfc] post-process the `maint print architecture' from gdb_mbuild.sh

[Andrew Cagney]

[-- Attachment #1: Type: text/plain, Size: 925 bytes --]

> On Tue, Jan 07, 2003 at 10:10:36AM -0500, Andrew Cagney wrote:
> 
>> >>Thoughts?
> 
>> >
>> >
>> >The idea sounds good to me.
>> >
>> >Instead of ed'ing Gdb.log, how about just naming the first file
>> >Gdb.out or Gdb-raw.log, and then sed'ing into Gdb.log.
> 
>> 
>> I can save the unedited output.
>> 
>> To make sed worthwhile, I'd need to construct a sed script that did all 
>> substitutions in a single pass.  Otherwize the script ends up repeating:
>> 	sed -e ... < input > output
>> 	mv output input
>> which is equivalent to ed.  That, I think, is something for a rainy day.
> 
> 
> It's as simple as "sed -e ... -e ... -e ... -e ... < input > output",
> isn't it?  Or via -f and a file, the same thing.

Yes - `construct a sed script'.  I've done it (attached, committed). 
Doesn't make it any faster though.  The slow bit is:

+       func="`addr2line -f -e ./gdb/gdb -s ${addr} | sed -n -e 1p`"

enjoy,
Andrew


[-- Attachment #2: diffs --]
[-- Type: text/plain, Size: 1589 bytes --]

2003-01-08  Andrew Cagney  <ac131313@redhat.com>

	* gdb_mbuild.sh: Edit the output of `maint print architecture'
	replacing hex constants with function names and stripping leading
	file name directory prefixes.

Index: gdb_mbuild.sh
===================================================================
RCS file: /cvs/src/src/gdb/gdb_mbuild.sh,v
retrieving revision 1.3
diff -u -r1.3 gdb_mbuild.sh
--- gdb_mbuild.sh	2 Jan 2003 16:40:33 -0000	1.3
+++ gdb_mbuild.sh	8 Jan 2003 23:46:52 -0000
@@ -279,12 +279,32 @@
     fail "gdb printed no output" ! -s Gdb.log
     grep -e internal-error Gdb.log && fail "gdb panic" 1
 
+    echo ... cleanup ${target}
+
+    # Create a sed script that cleans up the output from GDB.
+    rm -f mbuild.sed
+    touch mbuild.sed || exit 1
+
+    # Rules to replace <0xNNNN> with the corresponding function's
+    # name.
+    sed -n -e '/<0x0*>/d' -e 's/^.*<0x\([0-9a-f]*\)>.*$/0x\1/p' Gdb.log \
+    | sort -u \
+    | while read addr
+    do
+	func="`addr2line -f -e ./gdb/gdb -s ${addr} | sed -n -e 1p`"
+	test ${verbose} -gt 0 && echo "${addr} ${func}" 1>&2
+	echo "s/<${addr}>/<${func}>/g"
+    done >> mbuild.sed
+
+    # Rules to strip the leading paths off of file names.
+    echo 's/"\/.*\/gdb\//"gdb\//g' >> mbuild.sed
+
     # Replace the build directory with a file as semaphore that stops
     # a rebuild. (should the logs be saved?)
 
     cd ${builddir}
     rm -f ${target}.tmp
-    mv ${target}/Gdb.log ${target}.tmp
+    sed -f ${target}/mbuild.sed ${target}/Gdb.log > ${target}.tmp
     rm -rf ${target}
     mv ${target}.tmp ${target}
 

Re: [patch/rfc] post-process the `maint print architecture' from gdb_mbuild.sh

[Daniel Jacobowitz]

On Wed, Jan 08, 2003 at 06:59:10PM -0500, Andrew Cagney wrote:
> >On Tue, Jan 07, 2003 at 10:10:36AM -0500, Andrew Cagney wrote:
> >
> >>>>Thoughts?
> >
> >>>
> >>>
> >>>The idea sounds good to me.
> >>>
> >>>Instead of ed'ing Gdb.log, how about just naming the first file
> >>>Gdb.out or Gdb-raw.log, and then sed'ing into Gdb.log.
> >
> >>
> >>I can save the unedited output.
> >>
> >>To make sed worthwhile, I'd need to construct a sed script that did all 
> >>substitutions in a single pass.  Otherwize the script ends up repeating:
> >>	sed -e ... < input > output
> >>	mv output input
> >>which is equivalent to ed.  That, I think, is something for a rainy day.
> >
> >
> >It's as simple as "sed -e ... -e ... -e ... -e ... < input > output",
> >isn't it?  Or via -f and a file, the same thing.
> 
> Yes - `construct a sed script'.  I've done it (attached, committed). 
> Doesn't make it any faster though.  The slow bit is:

Thanks!

> +       func="`addr2line -f -e ./gdb/gdb -s ${addr} | sed -n -e 1p`"

Yeah.  Let me see if I can finish the other bit of code which has to
find the binary anyway; then we can try doing this internally in the
maint print architecture dump.  That'd do wonders for our eyes :)


-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: [patch/rfc] post-process the `maint print architecture' from gdb_mbuild.sh

[Michael Elizabeth Chastain]

> I can save the unedited output.

Works for me.

> To make sed worthwhile, I'd need to construct a sed script that did all 
> substitutions in a single pass.  Otherwize the script ends up repeating:
> 	sed -e ... < input > output
> 	mv output input
> which is equivalent to ed.  That, I think, is something for a rainy day.

I am fond of the "_P_rogram to replace _E_d scripts on _R_ainy-days
_L_anguage", myself.

Michael C

Re: [patch/rfc] post-process the `maint print architecture' from gdb_mbuild.sh

[Michael Elizabeth Chastain]

> It's as simple as "sed -e ... -e ... -e ... -e ... < input > output",
> isn't it?  Or via -f and a file, the same thing.

The problem is that the "-e ..." thingies are computed dynamically.
The "-f" idea would work though I think.

Michael C