Re: [PATCH, gdb/testsuite] Fix compare-sections.exp FAIL

[Thomas Preudhomme <thomas dot preudhomme at foss dot arm dot com>]

Hi Pedro,

Thanks for the review. Please find an updated patch attached. Updated ChangeLog 
entry and description are as follow:

compare-sections.exp has two cases that are not handled appropriately:
1) value read from read-only section is negative
2) error while patching that section

This patch adapts the regular expression to allow a minus sign to deal
with 1) and test for the error message to not set written if read-only
section cannot be written to so as to solve 2).

ChangeLog entry is as follows:

*** gdb/testsuite/ChangeLog ***

2017-09-11  Thomas Preud'homme  <thomas.preudhomme@arm.com>
	    Pedro Alves  <palves@redhat.com>

	* gdb.base/compare-sections.exp (get value of read-only section): Read
	as unsigned value.
	(corrupt read-only section): Don't set written if patching failed.

Without this patch I get a Tcl error on the first test while it now
PASSes and the second one is marked UNSUPPORTED.

Is this ok for master?

Best regards,

Thomas

On 06/09/17 18:13, Pedro Alves wrote:
> On 09/06/2017 05:58 PM, Thomas Preudhomme wrote:
>
> > diff --git a/gdb/testsuite/gdb.base/compare-sections.exp b/gdb/testsuite/gdb.base/compare-sections.exp
> > index e4b99216408f86745a890f80175e2f26c39303b5..8a0cf89a4c7c51299c59f59175865b5eb211f213 100644
> > --- a/gdb/testsuite/gdb.base/compare-sections.exp
> > +++ b/gdb/testsuite/gdb.base/compare-sections.exp
> > @@ -118,7 +118,7 @@ with_test_prefix "read-only" {
> >   
> >       set test "get value of read-only section"
> >       gdb_test_multiple "print /d *(unsigned char *) $ro_address" "$test" {
> > -	-re " = (\[0-9\]*).*$gdb_prompt $" {
> > +	-re " = (-?\[0-9\]+).*$gdb_prompt $" {
>
> This is reading bytes off of memory, as we can infer from
> use of "unsigned char".   The test is most probably using
> explicit /d to avoid printing the number as a character.
> I think it'd be clearer to change the "/d" to "/u", here
> and below too.
>
> Also, I think treating as negative integer would actually
> be buggy because if the value happens to be 0xff, then you
> get back -1, which would incorrectly match here:
>
>      if {$orig == -1} {
> 	untested "couldn't read address of read-only section"
> 	return -1
>      }
> >   	    set orig $expect_out(1,string)
> >   	    pass "$test"
> >   	}
> > @@ -136,6 +136,8 @@ with_test_prefix "read-only" {
> >       set written -1
> >       set test "corrupt read-only section"
> >       gdb_test_multiple "print /d *(unsigned char *) $ro_address = $patch" "$test" {
> > +	-re " = .*Cannot access memory at address $ro_address.*$gdb_prompt $" {
>
> Add:
>   	    pass "$test (cannot write)"
>
> > +	}
> >   	-re " = (\[0-9\]*).*$gdb_prompt $" {
> >   	    set written $expect_out(1,string)
> >   	    pass "$test"
>
> Thanks,
> Pedro Alves
diff --git a/gdb/testsuite/gdb.base/compare-sections.exp b/gdb/testsuite/gdb.base/compare-sections.exp
index e4b99216408f86745a890f80175e2f26c39303b5..90676b5b9872984af90e2b6b5e602a8287f8356f 100644
--- a/gdb/testsuite/gdb.base/compare-sections.exp
+++ b/gdb/testsuite/gdb.base/compare-sections.exp
@@ -117,7 +117,7 @@ with_test_prefix "read-only" {
     set orig -1
 
     set test "get value of read-only section"
-    gdb_test_multiple "print /d *(unsigned char *) $ro_address" "$test" {
+    gdb_test_multiple "print /u *(unsigned char *) $ro_address" "$test" {
 	-re " = (\[0-9\]*).*$gdb_prompt $" {
 	    set orig $expect_out(1,string)
 	    pass "$test"
@@ -136,6 +136,9 @@ with_test_prefix "read-only" {
     set written -1
     set test "corrupt read-only section"
     gdb_test_multiple "print /d *(unsigned char *) $ro_address = $patch" "$test" {
+	-re " = .*Cannot access memory at address $ro_address.*$gdb_prompt $" {
+	    pass "$test (cannot write)"
+	}
 	-re " = (\[0-9\]*).*$gdb_prompt $" {
 	    set written $expect_out(1,string)
 	    pass "$test"