possible bug with -fsecond-underscore ?

[Bernhard Reutner-Fischer]

Hi there.

gfc_sym_mangled_function_id() and gfc_sym_mangled_common_id both have
(identical, but that's for another patch to maybe factor it out) the
following code for underscore handling which looks suspicious:

---8<---
static tree
gfc_sym_mangled_common_id (gfc_common_head *com)
{
int has_underscore;
const char *name;

/* Get the name out of the common block pointer. */
name = com->name;

/* If we're supposed to do a bind(c). */
if (com->is_bind_c && com->binding_label)
return maybe_get_identifier (com->binding_label);

if (name == gfc_get_string (BLANK_COMMON_NAME))
return maybe_get_identifier (name);

if (flag_underscoring)
{
has_underscore = strchr (name, '_') != 0;
if (flag_second_underscore && has_underscore)
return gfc_get_identifier("%s__", name);
else
return gfc_get_identifier("%s_", name);
}
else
return maybe_get_identifier (name);
}
---8<---

the strchr looks somewhat suspicious, IMHO:
Isn't what the code really wants to check the fact whether or not the
name *already*ends* with an underscore?

***@h:/tmp/o$ printf "common /C/ j,i\nend" | gfortran -x f95-cpp-input -
-ffree-form -c -o huh.o -DC=foo && readelf -s huh.o | grep foo
14: 0000000000000010 8 OBJECT GLOBAL DEFAULT COM foo_
***@h:/tmp/o$ printf "common /C/ j,i\nend" | gfortran -x f95-cpp-input -
-ffree-form -c -o huh.o -DC=foo -fsecond-underscore && readelf -s
huh.o | grep foo
14: 0000000000000010 8 OBJECT GLOBAL DEFAULT COM foo_
***@h:/tmp/o$ printf "common /C/ j,i\nend" | gfortran -x f95-cpp-input -
-ffree-form -c -o huh.o -DC=foo_bar && readelf -s huh.o | grep foo
14: 0000000000000010 8 OBJECT GLOBAL DEFAULT COM foo_bar_
***@h:/tmp/o$ printf "common /C/ j,i\nend" | gfortran -x f95-cpp-input -
-ffree-form -c -o huh.o -DC=foo_bar -fsecond-underscore && readelf -s
huh.o | grep foo
14: 0000000000000010 8 OBJECT GLOBAL DEFAULT COM foo_bar__

I had expected "common /foo/" with -fsecond-underscore to produce a
"foo__" sym, no?
What am i missing?

thanks,

[Bernhard Reutner-Fischer]

On Mon, 24 Sep 2018 at 13:26, Bernhard Reutner-Fischer

Post by Bernhard Reutner-Fischer
Hi there.
gfc_sym_mangled_function_id() and gfc_sym_mangled_common_id both have
(identical, but that's for another patch to maybe factor it out) the
---8<---
static tree
gfc_sym_mangled_common_id (gfc_common_head *com)
{
int has_underscore;
const char *name;
/* Get the name out of the common block pointer. */
name = com->name;
/* If we're supposed to do a bind(c). */
if (com->is_bind_c && com->binding_label)
return maybe_get_identifier (com->binding_label);
if (name == gfc_get_string (BLANK_COMMON_NAME))
return maybe_get_identifier (name);
if (flag_underscoring)
{
has_underscore = strchr (name, '_') != 0;
if (flag_second_underscore && has_underscore)
return gfc_get_identifier("%s__", name);
else
return gfc_get_identifier("%s_", name);
}
else
return maybe_get_identifier (name);
}
---8<---
Isn't what the code really wants to check the fact whether or not the
name *already*ends* with an underscore?
-ffree-form -c -o huh.o -DC=foo && readelf -s huh.o | grep foo
14: 0000000000000010 8 OBJECT GLOBAL DEFAULT COM foo_
-ffree-form -c -o huh.o -DC=foo -fsecond-underscore && readelf -s
huh.o | grep foo
14: 0000000000000010 8 OBJECT GLOBAL DEFAULT COM foo_
-ffree-form -c -o huh.o -DC=foo_bar && readelf -s huh.o | grep foo
14: 0000000000000010 8 OBJECT GLOBAL DEFAULT COM foo_bar_
-ffree-form -c -o huh.o -DC=foo_bar -fsecond-underscore && readelf -s
huh.o | grep foo
14: 0000000000000010 8 OBJECT GLOBAL DEFAULT COM foo_bar__
I had expected "common /foo/" with -fsecond-underscore to produce a
"foo__" sym, no?
What am i missing?

I should have read the documentation of -fsecond-underscore for
starters and maybe PR51820
"[doc] underscoring documentation incorrect" .

sorry for the noise.. :(