Skip to contentSkip to conversation

move -Wno-error=uninitialized to configure

PATCH In review Michael S. Tsirkin 4 participants 8 messages Jun 30, 2009

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 move -Wno-error=uninitialized to configure

Move -Wno-error=uninitialized out of rules.mak and into configure. Only use it if supported by compiler.

---
 configure |   15 ++++++++++++++-
 rules.mak |    2 +-
 2 files changed, 15 insertions(+), 2 deletions(-)
configure +14 -1
index eb9d73a..69c7471 100755
@@ -526,8 +526,21 @@ if test -z "$werror" ; then
     fi
 fi
 
+# Disable uninitialized variable warning or convert it to non-error, if possible
+cat > $TMPC <<EOF
+int main(void) {}
+EOF
+if $cc -Werror -Wno-error=uninitialized -c -o $TMPO $TMPC > /dev/null 2> /dev/null ; then
+    # Enable werror by default for recent compilers
+    werror_cflags="-Werror -Wno-error=uninitialized"
+elif $cc -Werror -Wno-uninitialized -c -o $TMPO $TMPC > /dev/null 2> /dev/null ; then
+    werror_cflags="-Werror -Wno-uninitialized"
+else
+    werror_cflags="-Werror"
+fi
+
 if test "$werror" = "yes" ; then
-    CFLAGS="$CFLAGS -Werror"
+    CFLAGS="$CFLAGS $werror_cflags"
 fi
 
 if test "$solaris" = "no" ; then
rules.mak +1 -1
index defee1d..8d6d96e 100644
@@ -1,6 +1,6 @@
 
 %.o: %.c
-	$(call quiet-command,$(CC) $(CPPFLAGS) $(CFLAGS) -Werror -Wno-error=uninitialized -c -o $@ $<,"  CC    $(TARGET_DIR)$@")
+	$(call quiet-command,$(CC) $(CPPFLAGS) $(CFLAGS) -c -o $@ $<,"  CC    $(TARGET_DIR)$@")
 
 %.o: %.S
 	$(call quiet-command,$(CC) $(CPPFLAGS) -c -o $@ $<,"  AS    $(TARGET_DIR)$@")
Signature · 1 line
1.6.2.2
#1 lore raw
Quoted 2 lines · Michael S. Tsirkin

Move -Wno-error=uninitialized out of rules.mak and into configure. Only use it if supported by compiler.

Why? Shouldn't we be fixing the uninitialized variables?

Paul

#2 lore raw
Quoted 7 lines · Paul Brook
Quoted 2 lines · Michael S. Tsirkin

Move -Wno-error=uninitialized out of rules.mak and into configure. Only use it if supported by compiler.

Why? Shouldn't we be fixing the uninitialized variables?

Paul

gcc's algorithm for detecting uninitialized variables is famously unreliable. Just scattering '= 0' around the code masks the warning but hides the fact that application logic might still be broken.

Signature · 1 line
MST
#3 lore raw
Quoted 5 lines · Michael S. Tsirkin
Quoted 4 lines · Paul Brook
Quoted 2 lines · Michael S. Tsirkin

Move -Wno-error=uninitialized out of rules.mak and into configure. Only use it if supported by compiler.

Quoted 8 lines
Quoted 3 lines

Why? Shouldn't we be fixing the uninitialized variables?

Paul

gcc's algorithm for detecting uninitialized variables is famously unreliable. Just scattering '= 0' around the code masks the warning but hides the fact that application logic might still be broken.

To clarify: what's broken is that gcc has false positives in the logic. And if you work around this by adding initialization where it's not really required, you confuse other tools that would otherwise have a chance to detect an error.

Signature · 1 line
MST
#7 lore raw
In reply to Paul Brook
Quoted 8 lines · Paul Brook
Quoted 3 lines · Michael S. Tsirkin

Move -Wno-error=uninitialized out of rules.mak and into configure. Only use it if supported by compiler.

Why? Shouldn't we be fixing the uninitialized variables?

Fixing older compilers to support -Wno-error is out of the scope of this project.

Signature · 1 line
error compiling committee.c: too many arguments to function
#4 lore raw
In reply to Michael S. Tsirkin
Quoted 15 lines · Michael S. Tsirkin
Quoted 5 lines · Michael S. Tsirkin
Quoted 4 lines · Paul Brook
Quoted 2 lines · Michael S. Tsirkin

Move -Wno-error=uninitialized out of rules.mak and into configure. Only use it if supported by compiler.

Quoted 8 lines
Quoted 3 lines

Why? Shouldn't we be fixing the uninitialized variables?

Paul

gcc's algorithm for detecting uninitialized variables is famously unreliable. Just scattering '= 0' around the code masks the warning but hides the fact that application logic might still be broken.

Quoted 4 lines

To clarify: what's broken is that gcc has false positives in the logic. And if you work around this by adding initialization where it's not really required, you confuse other tools that would otherwise have a chance to detect an error.

In that case should we be using -Wno-uninitialized?

The whole point of -Werror is that it forces us to keep the code clean. Having to mentally filter out false positives just doesn't work. Once you have routinely expected warnings then people stop checking for new ones.

Paul

#5 lore raw
Quoted 21 lines · Paul Brook
Quoted 15 lines · Michael S. Tsirkin
Quoted 5 lines · Michael S. Tsirkin
Quoted 4 lines · Paul Brook
Quoted 2 lines · Michael S. Tsirkin

Move -Wno-error=uninitialized out of rules.mak and into configure. Only use it if supported by compiler.

Quoted 8 lines
Quoted 3 lines

Why? Shouldn't we be fixing the uninitialized variables?

Paul

gcc's algorithm for detecting uninitialized variables is famously unreliable. Just scattering '= 0' around the code masks the warning but hides the fact that application logic might still be broken.

Quoted 4 lines

To clarify: what's broken is that gcc has false positives in the logic. And if you work around this by adding initialization where it's not really required, you confuse other tools that would otherwise have a chance to detect an error.

Quoted 7 lines

In that case should we be using -Wno-uninitialized?

The whole point of -Werror is that it forces us to keep the code clean. Having to mentally filter out false positives just doesn't work. Once you have routinely expected warnings then people stop checking for new ones.

Paul

That's what -Wno-error=uninitialized does: makes the filtering automatically so you don't need to filter it mentally.

Signature · 1 line
MST
#6 lore raw
Quoted 7 lines · Michael S. Tsirkin
Quoted 3 lines · Paul Brook

The whole point of -Werror is that it forces us to keep the code clean. Having to mentally filter out false positives just doesn't work. Once you have routinely expected warnings then people stop checking for new ones.

That's what -Wno-error=uninitialized does: makes the filtering automatically so you don't need to filter it mentally.

Right, but not all versions of GCC that support '-Werror' support '-Wno-error=...'. As I tried to hint at previously, the only sensible way to deal with this is to use '-Werror=...' to be specific about the types of checking that you want, so that you can avoid checks that produce false positives, and only enable the checking on versions of GCC that support it. Otherwise, future versions of GCC could produce all sorts of false positives that you have no control over. '-Werror=...' doesn't fix that entirely, but at least it's a start.

I've already seen several 'fixes' to QEMU to suppress false positives. For some strange reason people always pick the magic value '0' when 'shutting GCC up'. *sigh* At the very least, *please please please* consider something like Linux's uninitialized_var() macro for this.

(BTW, recommended reading: http://lwn.net/Articles/331593/)

Cheers,

Signature · 1 line
Stuart Brady