c: Propagate erroneous types to declaration specifiers [PR107805]

Message ID 87cz9fqixe.fsf@oldenburg.str.redhat.com
State Unresolved
Headers
Series c: Propagate erroneous types to declaration specifiers [PR107805] |

Checks

Context Check Description
snail/gcc-patch-check warning Git am fail log

Commit Message

Florian Weimer Nov. 22, 2022, 12:21 p.m. UTC
  Without this change, finish_declspecs cannot tell that whether there
was an erroneous type specified, or no type at all.  This may result
in additional diagnostics for implicit ints, or missing diagnostics
for multiple types.

	PR c/107805

gcc/c/
	* c-decl.cc (declspecs_add_type): Propagate error_mark_bode
	from type to specs.

gcc/testsuite/
	* gcc.dg/pr107805-1.c: New test.
	* gcc.dg/pr107805-1.c: Likewise.

---
Note regarding testing: I boostrap with c,c++,lto on x86-64
(non-multlib) and diffed these .sum files:

gcc/testsuite/gcc/gcc.sum
gcc/testsuite/g++/g++.sum
x86_64-pc-linux-gnu/libgomp/testsuite/libgomp.sum
x86_64-pc-linux-gnu/libstdc++-v3/testsuite/libstdc++.sum
x86_64-pc-linux-gnu/libatomic/testsuite/libatomic.sum
x86_64-pc-linux-gnu/libitm/testsuite/libitm.sum

Apart from timestamps, the only differences I get is this change:


base-commit: e4faee8d02ec5d65bf418612f7181823eb08c078
  

Comments

Joseph Myers Nov. 22, 2022, 11:05 p.m. UTC | #1
On Tue, 22 Nov 2022, Florian Weimer via Gcc-patches wrote:

> Without this change, finish_declspecs cannot tell that whether there
> was an erroneous type specified, or no type at all.  This may result
> in additional diagnostics for implicit ints, or missing diagnostics
> for multiple types.
> 
> 	PR c/107805
> 
> gcc/c/
> 	* c-decl.cc (declspecs_add_type): Propagate error_mark_bode
> 	from type to specs.
> 
> gcc/testsuite/
> 	* gcc.dg/pr107805-1.c: New test.
> 	* gcc.dg/pr107805-1.c: Likewise.

OK.
  
Florian Weimer Nov. 24, 2022, 10:01 a.m. UTC | #2
* Joseph Myers:

> On Tue, 22 Nov 2022, Florian Weimer via Gcc-patches wrote:
>
>> Without this change, finish_declspecs cannot tell that whether there
>> was an erroneous type specified, or no type at all.  This may result
>> in additional diagnostics for implicit ints, or missing diagnostics
>> for multiple types.
>> 
>> 	PR c/107805
>> 
>> gcc/c/
>> 	* c-decl.cc (declspecs_add_type): Propagate error_mark_bode
>> 	from type to specs.
>> 
>> gcc/testsuite/
>> 	* gcc.dg/pr107805-1.c: New test.
>> 	* gcc.dg/pr107805-1.c: Likewise.
>
> OK.

Thanks.  Permission to backport this to GCC 12 after a week or two?

Florian
  
Jakub Jelinek Nov. 24, 2022, 10:28 a.m. UTC | #3
On Thu, Nov 24, 2022 at 11:01:40AM +0100, Florian Weimer via Gcc-patches wrote:
> * Joseph Myers:
> 
> > On Tue, 22 Nov 2022, Florian Weimer via Gcc-patches wrote:
> >
> >> Without this change, finish_declspecs cannot tell that whether there
> >> was an erroneous type specified, or no type at all.  This may result
> >> in additional diagnostics for implicit ints, or missing diagnostics
> >> for multiple types.
> >> 
> >> 	PR c/107805
> >> 
> >> gcc/c/
> >> 	* c-decl.cc (declspecs_add_type): Propagate error_mark_bode
> >> 	from type to specs.
> >> 
> >> gcc/testsuite/
> >> 	* gcc.dg/pr107805-1.c: New test.
> >> 	* gcc.dg/pr107805-1.c: Likewise.
> >
> > OK.
> 
> Thanks.  Permission to backport this to GCC 12 after a week or two?

In this case I'd wait a month, it will take some time until possible
error recovery bugs are discovered.

	Jakub
  
Florian Weimer Nov. 24, 2022, 10:53 a.m. UTC | #4
* Jakub Jelinek:

> On Thu, Nov 24, 2022 at 11:01:40AM +0100, Florian Weimer via Gcc-patches wrote:
>> * Joseph Myers:
>> 
>> > On Tue, 22 Nov 2022, Florian Weimer via Gcc-patches wrote:
>> >
>> >> Without this change, finish_declspecs cannot tell that whether there
>> >> was an erroneous type specified, or no type at all.  This may result
>> >> in additional diagnostics for implicit ints, or missing diagnostics
>> >> for multiple types.
>> >> 
>> >> 	PR c/107805
>> >> 
>> >> gcc/c/
>> >> 	* c-decl.cc (declspecs_add_type): Propagate error_mark_bode
>> >> 	from type to specs.
>> >> 
>> >> gcc/testsuite/
>> >> 	* gcc.dg/pr107805-1.c: New test.
>> >> 	* gcc.dg/pr107805-1.c: Likewise.
>> >
>> > OK.
>> 
>> Thanks.  Permission to backport this to GCC 12 after a week or two?
>
> In this case I'd wait a month, it will take some time until possible
> error recovery bugs are discovered.

Okay, I have made a note to backport it in the new year.  Hopefully
any regressions will be flagged on the PR or linked to it.

Thanks,
Florian
  

Patch

--- ./gcc/testsuite/gcc/gcc.sum 2022-11-22 05:45:33.813264761 -0500
+++ /tmp/b/build/./gcc/testsuite/gcc/gcc.sum    2022-11-22 06:39:10.667590185 -0500
@@ -83303,6 +83303,11 @@ 
 PASS: gcc.dg/pr107618.c  (test for bogus messages, line 9)
 PASS: gcc.dg/pr107618.c (test for excess errors)
 PASS: gcc.dg/pr107686.c (test for excess errors)
+PASS: gcc.dg/pr107805-1.c  (test for errors, line 3)
+PASS: gcc.dg/pr107805-1.c (test for excess errors)
+PASS: gcc.dg/pr107805-2.c  (test for errors, line 3)
+PASS: gcc.dg/pr107805-2.c  (test for errors, line 4)
+PASS: gcc.dg/pr107805-2.c (test for excess errors)
 PASS: gcc.dg/pr11459-1.c (test for excess errors)
 PASS: gcc.dg/pr11492.c  (test for bogus messages, line 8)
 PASS: gcc.dg/pr11492.c (test for excess errors)
@@ -190486,7 +190491,7 @@ 
 
                === gcc Summary ===
 
-# of expected passes           185932
+# of expected passes           185937
 # of unexpected failures       99
 # of unexpected successes      20
 # of expected failures         1484

So I think this means there are no test suite regressions.

Thanks,
Florian

 gcc/c/c-decl.cc                   | 6 ++----
 gcc/testsuite/gcc.dg/pr107805-1.c | 5 +++++
 gcc/testsuite/gcc.dg/pr107805-2.c | 4 ++++
 3 files changed, 11 insertions(+), 4 deletions(-)

diff --git a/gcc/c/c-decl.cc b/gcc/c/c-decl.cc
index 098e475f65d..4adb89e4aaf 100644
--- a/gcc/c/c-decl.cc
+++ b/gcc/c/c-decl.cc
@@ -12243,11 +12243,9 @@  declspecs_add_type (location_t loc, struct c_declspecs *specs,
     error_at (loc, "two or more data types in declaration specifiers");
   else if (TREE_CODE (type) == TYPE_DECL)
     {
-      if (TREE_TYPE (type) == error_mark_node)
-	; /* Allow the type to default to int to avoid cascading errors.  */
-      else
+      specs->type = TREE_TYPE (type);
+      if (TREE_TYPE (type) != error_mark_node)
 	{
-	  specs->type = TREE_TYPE (type);
 	  specs->decl_attr = DECL_ATTRIBUTES (type);
 	  specs->typedef_p = true;
 	  specs->explicit_signed_p = C_TYPEDEF_EXPLICITLY_SIGNED (type);
diff --git a/gcc/testsuite/gcc.dg/pr107805-1.c b/gcc/testsuite/gcc.dg/pr107805-1.c
new file mode 100644
index 00000000000..559b6a5586e
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/pr107805-1.c
@@ -0,0 +1,5 @@ 
+/* { dg-do compile } */
+typedef int t;
+typedef struct { double a; int b; } t; /* { dg-error "conflicting types" } */
+t x; /* No warning here.  */
+
diff --git a/gcc/testsuite/gcc.dg/pr107805-2.c b/gcc/testsuite/gcc.dg/pr107805-2.c
new file mode 100644
index 00000000000..fa5fa4ce273
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/pr107805-2.c
@@ -0,0 +1,4 @@ 
+/* { dg-do compile } */
+typedef int t;
+typedef struct { double a; int b; } t; /* { dg-error "conflicting types" } */
+t char x; /* { dg-error "two or more data types" } */