Re: [PATCH 00/30] Rewrite the DWARF "partial" reader

[Tom de Vries <tdevries@suse.de>]

On 8/26/21 11:29 PM, Tom Tromey wrote:
> These all pass for me:
> 
> Tom> ERROR: Couldn't load cpexprs-debug-types into GDB (eof).
> 
> # of expected passes		393
> 

This fails intermittently for me, so that could indeed be a
thread-safety bug (using worker_threads=1/unlimited):
...
$ total=100; worker_threads=1; pass=0; for n in $(seq 1 $total); do gdb
-q -batch -iex "maint set worker-threads $worker_threads"
./outputs/gdb.cp/cpexprs-debug-types/cpexprs-debug-types >LOG 2>ERR; if
[ $? -eq 0 ]; then pass=$(($pass + 1)); fi; done; echo "PASS: $pass/$total"
PASS: 100/100
$ total=100; worker_threads=unlimited; pass=0; for n in $(seq 1 $total);
do gdb -q -batch -iex "maint set worker-threads $worker_threads"
./outputs/gdb.cp/cpexprs-debug-types/cpexprs-debug-types >LOG 2>ERR; if
[ $? -eq 0 ]; then pass=$(($pass + 1)); fi; done; echo "PASS: $pass/$total"
PASS: 62/100
...

The run above is with gcc-7, using gcc-11 I ran twice and got "PASS:
96/100" and "PASS: 100/100".  So, perhaps less likely to fail with newer
gcc?

> Tom> FAIL: gdb.dwarf2/dw2-using-debug-str.exp: file dw2-using-debug-str
> 
> # of expected passes		3
> 

This fails reliably for me, with gcc-7 - gcc-11.

I think basically the problem is that we get:
...
(gdb) file
/home/vries/gdb_versions/devel/build/gdb/testsuite/outputs/gdb.dwarf2/dw2-using-debug-str/dw2-using-debug-str-no-debug-str^M
Reading symbols from
/home/vries/gdb_versions/devel/build/gdb/testsuite/outputs/gdb.dwarf2/dw2-using-debug-str/dw2-using-debug-str-no-debug-str...^M
Dwarf Error: DW_FORM_strp used without required section^M
Dwarf Error: DW_FORM_strp used without required section^M
Dwarf Error: DW_FORM_strp used without required section^M
Dwarf Error: DW_FORM_strp used without required section^M
Dwarf Error: DW_FORM_strp used without required section^M
Dwarf Error: DW_FORM_strp used without required section^M
(gdb) FAIL: gdb.dwarf2/dw2-using-debug-str.exp: file dw2-using-debug-str
...
while the test-case expects just one dwarf error.  I don't know whether
that should be solved by only emitting one dwarf error, or updating the
test-case.

I don't understand though why you don't see this.  Is the threading
somehow not activated in that test-case?  Or is this some timing issue
where the first dwarf error cancels the other threads before they get to
emit the other dwarf errors?

> Tom> FAIL: gdb.dwarf2/fission-multi-cu.exp: ptype func
> Tom> FAIL: gdb.dwarf2/fission-multi-cu.exp: break func
> Tom> FAIL: gdb.dwarf2/fission-multi-cu.exp: continue to func
> Tom> FAIL: gdb.dwarf2/fission-multi-cu.exp: frame in func
> 
> # of expected passes		6

This fails intermittently for me, so that could indeed be a
thread-safety bug (using worker_threads=1/unlimited):
...
$ total=100; worker_threads=1; pass=0; for n in $(seq 1 $total); do gdb
-q -batch -iex "maint set worker-threads $worker_threads"
outputs/gdb.dwarf2/fission-multi-cu/fission-multi-cu -ex "ptype func"
>LOG 2>ERR; if grep -q 'type = int (int)' LOG; then pass=$(($pass + 1));
fi; done; echo "PASS: $pass/$total"
PASS: 100/100
$ total=100; worker_threads=unlimited; pass=0; for n in $(seq 1 $total);
do gdb -q -batch -iex "maint set worker-threads $worker_threads"
outputs/gdb.dwarf2/fission-multi-cu/fission-multi-cu -ex "ptype func"
>LOG 2>ERR; if grep -q 'type = int (int)' LOG; then pass=$(($pass + 1));
fi; done; echo "PASS: $pass/$total"
*** Error in `/home/vries/gdb_versions/devel/gdb': malloc(): memory
corruption: 0x00007fe544009790 ***
*** Error in `/home/vries/gdb_versions/devel/gdb': free(): invalid
pointer: 0x00007fe544000994 ***
PASS: 95/100
...

The run above is with gcc-7, with gcc-11 I got: "PASS: 97/100" and
"PASS: 95/100", so that's pretty similar.

Thanks,
- Tom

> 
> 
> Maybe it's compiler dependent?
> Or perhaps some kind of thread-safety bug.
> 
> Tom
>