Re: [PATCHES] fix for strict-alias warnings
"Andrew Dunstan" <[EMAIL PROTECTED]> writes: > Interestingly, a lot of the comparison and call to memset() still seem to be > optimised away, but the loop from MemSet is left, so the multiplication is > also not optimised away. Yeah, I saw the same in gcc 3.2 (on Intel) yesterday. I thought maybe 3.3 would fix that, since it sure looks like an optimization bug, but evidently not. regards, tom lane ---(end of broadcast)--- TIP 4: Don't 'kill -9' the postmaster
Re: [PATCHES] fix for strict-alias warnings
Andrew Dunstan <[EMAIL PROTECTED]> writes: > All this is interesting, but the real problem remains that we don't know > what else might be affected because gcc apparently doesn't promise to > tell us. IMO the gcc team made a bad mistake by turning this on by > default for -O2 without reliable accompanying diagnostics. Yeah, this seems to be the killer point. It *might* be safe to leave -fstrict-aliasing on; but we can't tell. I think we can safely just make the default switches for gcc be CFLAGS="-O2 -fno-strict-aliasing". If I understand correctly, this will work with all gcc versions back to 2.7.something, which is pretty much ancient history now --- and if anyone needs to make it work with such an old gcc, all they need do is specify CFLAGS to configure instead of letting it default. Will commit this change unless I hear objections soon. regards, tom lane ---(end of broadcast)--- TIP 1: subscribe and unsubscribe commands go to [EMAIL PROTECTED]
Re: [PATCHES] fix for strict-alias warnings
gcc 3.3.1/cygwin MemSetAligned: prints "square is 4.00" MemSet: prints "square is 0.00" Interestingly, a lot of the comparison and call to memset() still seem to be optimised away, but the loop from MemSet is left, so the multiplication is also not optimised away. here's the assembler for test2/MemSet: subl$28, %esp leal16(%esp), %eax movl$0, 16(%esp) leal24(%esp), %edx cmpl%edx, %eax movl$1073741824, 20(%esp) jae L21 .align 16 L26: movl$0, (%eax) addl$4, %eax cmpl%edx, %eax jb L26 L21: fldl16(%esp) movl$LC1, (%esp) fmul%st(0), %st fstl16(%esp) fstpl 4(%esp) call_printf addl$28, %esp ret cheers andrew - Original Message - From: "Manfred Spraul" <[EMAIL PROTECTED]> To: "Tom Lane" <[EMAIL PROTECTED]> Cc: "Andrew Dunstan" <[EMAIL PROTECTED]>; "Patches (PostgreSQL)" <[EMAIL PROTECTED]> Sent: Tuesday, October 14, 2003 5:01 PM Subject: Re: [PATCHES] fix for strict-alias warnings > Tom Lane wrote: > > >Manfred Spraul <[EMAIL PROTECTED]> writes: > > > > > >>After some massaging, I've succeeded in generating bad code using a > >>slightly modified MemSetAligned macro (parameters -O2 > >>-fstrict-aliasing): gcc pipelined the x*x around the memset. > >> > >> > > > >As I already explained, we do not care about the MemSetAligned case. > >Is gcc 3.3 smart enough to optimize away the pointer alignment test > >in the full macro? > > > 3.2 optimizes away the pointer alignment test, but then doesn't pipeline > the "x*x" calculation. It might be due to a known (and now fixed) bug in > gcc where is lost track of constants, and thus didn't succeed in > optimizing long calculations. > > I don't have gcc 3.3 installed, but IMHO it would be insane to leave > strict alias analysis enabled - writing to *(int32*)addr violates the > alias rules, the bad code generated with MemSetAligned proved that. > Is someone around with 3.3 who could test MemSet? > > -- > Manfred > ---(end of broadcast)--- TIP 4: Don't 'kill -9' the postmaster
Re: [PATCHES] fix for strict-alias warnings
Manfred Spraul <[EMAIL PROTECTED]> writes: > Tom Lane wrote: >> Is gcc 3.3 smart enough to optimize away the pointer alignment test >> in the full macro? >> > 3.2 optimizes away the pointer alignment test, but then doesn't pipeline > the "x*x" calculation. Hm, confirmed here. So indeed it seems that Bruce was on the right track --- setting up a Node structure as a local variable may be a contributing factor. > I don't have gcc 3.3 installed, but IMHO it would be insane to leave > strict alias analysis enabled - writing to *(int32*)addr violates the > alias rules, the bad code generated with MemSetAligned proved that. While I don't really disagree, I am curious as to whether we are actually forestalling any bugs; so far I'm not convinced that the reported warnings correspond to real risks ... regards, tom lane ---(end of broadcast)--- TIP 6: Have you searched our list archives? http://archives.postgresql.org
Re: [PATCHES] fix for strict-alias warnings
Manfred Spraul wrote: Tom Lane wrote: Manfred Spraul <[EMAIL PROTECTED]> writes: After some massaging, I've succeeded in generating bad code using a slightly modified MemSetAligned macro (parameters -O2 -fstrict-aliasing): gcc pipelined the x*x around the memset. As I already explained, we do not care about the MemSetAligned case. Is gcc 3.3 smart enough to optimize away the pointer alignment test in the full macro? 3.2 optimizes away the pointer alignment test, but then doesn't pipeline the "x*x" calculation. It might be due to a known (and now fixed) bug in gcc where is lost track of constants, and thus didn't succeed in optimizing long calculations. I don't have gcc 3.3 installed, but IMHO it would be insane to leave strict alias analysis enabled - writing to *(int32*)addr violates the alias rules, the bad code generated with MemSetAligned proved that. Is someone around with 3.3 who could test MemSet? I have it on my cygwin installation - that's how this whole mess arose in the first place ;-) All this is interesting, but the real problem remains that we don't know what else might be affected because gcc apparently doesn't promise to tell us. IMO the gcc team made a bad mistake by turning this on by default for -O2 without reliable accompanying diagnostics. cheers andrew ---(end of broadcast)--- TIP 5: Have you checked our extensive FAQ? http://www.postgresql.org/docs/faqs/FAQ.html
Re: [PATCHES] fix for strict-alias warnings
Tom Lane wrote: Manfred Spraul <[EMAIL PROTECTED]> writes: After some massaging, I've succeeded in generating bad code using a slightly modified MemSetAligned macro (parameters -O2 -fstrict-aliasing): gcc pipelined the x*x around the memset. As I already explained, we do not care about the MemSetAligned case. Is gcc 3.3 smart enough to optimize away the pointer alignment test in the full macro? 3.2 optimizes away the pointer alignment test, but then doesn't pipeline the "x*x" calculation. It might be due to a known (and now fixed) bug in gcc where is lost track of constants, and thus didn't succeed in optimizing long calculations. I don't have gcc 3.3 installed, but IMHO it would be insane to leave strict alias analysis enabled - writing to *(int32*)addr violates the alias rules, the bad code generated with MemSetAligned proved that. Is someone around with 3.3 who could test MemSet? -- Manfred ---(end of broadcast)--- TIP 5: Have you checked our extensive FAQ? http://www.postgresql.org/docs/faqs/FAQ.html
Re: [PATCHES] fix for strict-alias warnings
Neil Conway writes: > Oh? Which platforms are you referring to, and what tests were performed? http://archives.postgresql.org/pgsql-patches/2002-10/msg00085.php -- Peter Eisentraut [EMAIL PROTECTED] ---(end of broadcast)--- TIP 3: if posting/reading through Usenet, please send an appropriate subscribe-nomail command to [EMAIL PROTECTED] so that your message can get through to the mailing list cleanly
Re: [PATCHES] fix for strict-alias warnings
Manfred Spraul <[EMAIL PROTECTED]> writes: > After some massaging, I've succeeded in generating bad code using a > slightly modified MemSetAligned macro (parameters -O2 > -fstrict-aliasing): gcc pipelined the x*x around the memset. As I already explained, we do not care about the MemSetAligned case. Is gcc 3.3 smart enough to optimize away the pointer alignment test in the full macro? (Hm, if so, maybe that explains Bruce's observation that the warning only shows up with Node structs that are allocated as local variables, rather than palloc'd ... that would be the only case where gcc could possibly optimize away the alignment test ...) regards, tom lane ---(end of broadcast)--- TIP 8: explain analyze is your friend
Re: [PATCHES] fix for strict-alias warnings
On Tue, 2003-10-14 at 16:29, Peter Eisentraut wrote: > The last time I did some testing, the builtin memset() was significantly > faster on plenty of different platforms. Oh? Which platforms are you referring to, and what tests were performed? You can find the benchmark results I'm referring to in the archives here: http://archives.postgresql.org/pgsql-hackers/2002-08/msg02116.php (In some hypothetical world in which MemSet() didn't offer a significant performance improvement, there is no reason to keep it around.) -Neil ---(end of broadcast)--- TIP 7: don't forget to increase your free space map settings
Re: [PATCHES] fix for strict-alias warnings
Tom Lane wrote:
Given that gcc is smart enough not to move any code across the memset()
call, I doubt that it would be moving anything across the whole if()
construct. Now if the if-condition were such that the memset code path
could be optimized away, then we'd have a problem, but in practice I do
not believe gcc is smart enough to realize that the alignment check is
always true.
gcc-3.2.2 optimizes the memset away - that's a simple exercise for gcc.
gcc-3.2.2 isn't smart enough to replace everything - it didn't like the
pointer arithmetics.
After some massaging, I've succeeded in generating bad code using a
slightly modified MemSetAligned macro (parameters -O2
-fstrict-aliasing): gcc pipelined the x*x around the memset.
Annotated asm output with gcc -O99 -fomit-frame-pointer -fstrict-aliasing:
08048328 :
8048328: 83 ec 18sub$0x18,%esp
> stack setup for automatic variables.
804832b: c7 44 24 0c 00 00 00movl $0x0,0xc(%esp,1)
8048332: 00
8048333: c7 44 24 10 00 00 00movl $0x4000,0x10(%esp,1)
804833a: 40
x = 1.0;
804833b: dd 44 24 0c fldl 0xc(%esp,1)
804833f: d8 c8 fmul %st(0),%st
x = x*x;
8048341: c7 44 24 0c 00 00 00movl $0x0,0xc(%esp,1)
8048348: 00
8048349: c7 44 24 10 00 00 00movl $0x0,0x10(%esp,1)
8048350: 00
> MemSetAligned(): optimized to storing two ints.
8048351: dd 54 24 0c fstl 0xc(%esp,1)
>>> write back the result of x*x to the stack
8048355: dd 1c 24fstpl (%esp,1)
>>> push x*x for printf call
8048358: 68 54 84 04 08 push $0x8048454
>>> push pointer to "square is %f.\n"
804835d: e8 06 ff ff ff call 8048268 <_init+0x38>
call printf
8048362: 83 c4 1cadd$0x1c,%esp
8048365: c3 ret
> and exit.
8048366: 89 f6 mov%esi,%esi
To paraphrase the ISO C line: gcc is not your grandfather's gcc. It's
within 10% of the best compilers for SpecInt - the propagation and
analysis of constants it quite good, and several bugs were fixed sinced
3.2.2.
What is the oldest gcc versions still supported by postgres? It seems
that the strict alias analysis is from the egcs tree. Probably first
supported by egcs-1.1.2 - is that gcc-2.91?
http://groups.google.de/groups?q=g:thl2087564510d&dq=&hl=de&lr=&ie=UTF-8&oe=UTF-8&selm=fa.fjlldvv.l7m7hk%40ifi.uio.no
--
Manfred
#include
#include
typedef signed int int32;
typedef int Size;
#define INT_ALIGN_MASK (sizeof(int)-1)
#define MEMSET_LOOP_LIMIT 1024
#define MemSetAligned(start, val, len) \
do \
{ \
int32 * _start = (int32 *) (start); \
int _val = (val); \
Size _len = (len); \
\
if ((_len & INT_ALIGN_MASK) == 0 && \
_val == 0 && \
_len <= MEMSET_LOOP_LIMIT) \
{ \
Sizeoffset; \
_len = _len/sizeof(int); \
for (offset=0;offset<_len;offset++) \
_start[offset] = 0; \
} \
else \
memset_is_this_function_optimized_away_question_mark((char *) _start, _val, _len); \
} while (0)
void test2(void)
{
double x;
x = 2.0;
MemSetAligned(&x, 0, sizeof(double));
x = x*x;
printf("square is %f.\n", x);
}
int main(void)
{
test2();
return 1;
}
---(end of broadcast)---
TIP 6: Have you searched our list archives?
http://archives.postgresql.org
Re: [PATCHES] fix for strict-alias warnings
Neil Conway writes: > On Tue, 2003-10-14 at 15:00, Manfred Spraul wrote: > > I think we must either add -fno-strict-aliasing, or switch to the c > > compiler memset functions for gcc. > > The last time we did some benchmarking, using the builtin memset() > imposes a significant performance penalty on plenty of different > platforms. The last time I did some testing, the builtin memset() was significantly faster on plenty of different platforms. -- Peter Eisentraut [EMAIL PROTECTED] ---(end of broadcast)--- TIP 2: you can get off all lists at once with the unregister command (send "unregister YourEmailAddressHere" to [EMAIL PROTECTED])
Re: [PATCHES] fix for strict-alias warnings
Tom Lane wrote: Bruce Momjian <[EMAIL PROTECTED]> writes: OK, what gcc versions support -fno-strict-aliasing? Do we need a configure test for it? Perhaps ... although it is recognized in 2.95.3 and probably for a good ways before that. It looks to me like what has changed in gcc 3.3 is not the existence of the flag, but the fact that -O2 now turns it on where it did not before. Yes, exactly - I have just confirmed this on the archived manuals on the gcc site. But turning it off for earlier releases does no harm. It is known to break 2.7.1 and I believe it will work with 2.95 and later. cheers andrew ---(end of broadcast)--- TIP 2: you can get off all lists at once with the unregister command (send "unregister YourEmailAddressHere" to [EMAIL PROTECTED])
Re: [PATCHES] fix for strict-alias warnings
Peter Eisentraut <[EMAIL PROTECTED]> writes: > Tom Lane writes: >> Given that gcc is smart enough not to move any code across the memset() >> call, > Is it? It had better be. regards, tom lane ---(end of broadcast)--- TIP 8: explain analyze is your friend
Re: [PATCHES] fix for strict-alias warnings
Tom Lane writes: > Given that gcc is smart enough not to move any code across the memset() > call, Is it? If you violate the aliasing rules, all bets are off. -- Peter Eisentraut [EMAIL PROTECTED] ---(end of broadcast)--- TIP 7: don't forget to increase your free space map settings
Re: [PATCHES] fix for strict-alias warnings
Tom Lane wrote: > Bruce Momjian <[EMAIL PROTECTED]> writes: > > OK, what gcc versions support -fno-strict-aliasing? Do we need a > > configure test for it? > > Perhaps ... although it is recognized in 2.95.3 and probably for a good > ways before that. > > It looks to me like what has changed in gcc 3.3 is not the existence > of the flag, but the fact that -O2 now turns it on where it did not > before. Right. I am just not sure how old our gcc compilers are in the field. We can do it unconditionally and wait for a failure report. -- Bruce Momjian| http://candle.pha.pa.us [EMAIL PROTECTED] | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup.| Newtown Square, Pennsylvania 19073 ---(end of broadcast)--- TIP 2: you can get off all lists at once with the unregister command (send "unregister YourEmailAddressHere" to [EMAIL PROTECTED])
Re: [PATCHES] fix for strict-alias warnings
Bruce Momjian <[EMAIL PROTECTED]> writes: > OK, what gcc versions support -fno-strict-aliasing? Do we need a > configure test for it? Perhaps ... although it is recognized in 2.95.3 and probably for a good ways before that. It looks to me like what has changed in gcc 3.3 is not the existence of the flag, but the fact that -O2 now turns it on where it did not before. regards, tom lane ---(end of broadcast)--- TIP 4: Don't 'kill -9' the postmaster
Re: [PATCHES] fix for strict-alias warnings
Tom Lane wrote: > > I think we must either add -fno-strict-aliasing, or switch to the c > > compiler memset functions for gcc. > > We will not be doing the latter, for certain. OK, what gcc versions support -fno-strict-aliasing? Do we need a configure test for it? Would someone profile PostgreSQL with -fno-strict-aliasing and see if you can see a larger performance hit? -- Bruce Momjian| http://candle.pha.pa.us [EMAIL PROTECTED] | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup.| Newtown Square, Pennsylvania 19073 ---(end of broadcast)--- TIP 3: if posting/reading through Usenet, please send an appropriate subscribe-nomail command to [EMAIL PROTECTED] so that your message can get through to the mailing list cleanly
Re: [PATCHES] fix for strict-alias warnings
Manfred Spraul <[EMAIL PROTECTED]> writes:
> I've asked the question on the gcc devel list. The first reply was that
> MemSet violates strict aliasing rules:
No doubt it does, but that is not really the issue here; the issue IMHO
is whether there is any real risk involved. Remember that the macro is
really of the form
if (blah blah)
{
// unsafe code is here
}
else
{
memset(...);
}
Given that gcc is smart enough not to move any code across the memset()
call, I doubt that it would be moving anything across the whole if()
construct. Now if the if-condition were such that the memset code path
could be optimized away, then we'd have a problem, but in practice I do
not believe gcc is smart enough to realize that the alignment check is
always true.
We do have to be wary of MemSetAligned and MemSetLoop, but these are
only used in constrained places (routines that do nothing else with
the memory in question) so I think they are not a problem.
> I think we must either add -fno-strict-aliasing, or switch to the c
> compiler memset functions for gcc.
We will not be doing the latter, for certain.
regards, tom lane
---(end of broadcast)---
TIP 7: don't forget to increase your free space map settings
Re: [PATCHES] fix for strict-alias warnings
On Tue, 2003-10-14 at 15:00, Manfred Spraul wrote: > I think we must either add -fno-strict-aliasing, or switch to the c > compiler memset functions for gcc. The last time we did some benchmarking, using the builtin memset() imposes a significant performance penalty on plenty of different platforms. -Neil ---(end of broadcast)--- TIP 8: explain analyze is your friend
Re: [PATCHES] fix for strict-alias warnings
Manfred Spraul wrote: I've asked the question on the gcc devel list. The first reply was that MemSet violates strict aliasing rules: http://gcc.gnu.org/ml/gcc/2003-10/msg00524.html I think we must either add -fno-strict-aliasing, or switch to the c compiler memset functions for gcc. The concensus appears to be -fno-strict-aliasing cheers andrew ---(end of broadcast)--- TIP 8: explain analyze is your friend
Re: [PATCHES] fix for strict-alias warnings
Tom Lane wrote: > "Andrew Dunstan" <[EMAIL PROTECTED]> writes: > > Of course, the linux kernel is aimed at a limited set of compilers - as I > > understand it basically gcc although it has been made to build with Intel > > compilers - which makes things somewhat easier for them. What is our target > > set of compilers? What is our target version of C? > > "Pretty much anything that speaks ANSI C" is my usual feeling about > that. As yet we have not heard of any non-gcc compilers in which this > is a problem, although you have a point that some compiler somewhere may > do this and not have a way to turn it off :-( I now understand the gcc issues. Sorry for getting confused. I have removed the void* cast from MemSet until we understand this issue better. -- Bruce Momjian| http://candle.pha.pa.us [EMAIL PROTECTED] | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup.| Newtown Square, Pennsylvania 19073 ---(end of broadcast)--- TIP 8: explain analyze is your friend
Re: [PATCHES] fix for strict-alias warnings
I've asked the question on the gcc devel list. The first reply was that MemSet violates strict aliasing rules: http://gcc.gnu.org/ml/gcc/2003-10/msg00524.html I think we must either add -fno-strict-aliasing, or switch to the c compiler memset functions for gcc. -- Manfred ---(end of broadcast)--- TIP 2: you can get off all lists at once with the unregister command (send "unregister YourEmailAddressHere" to [EMAIL PROTECTED])
Re: [PATCHES] fix for strict-alias warnings
Peter Eisentraut wrote: Andrew Dunstan writes: And looking at it even closer, 2 of the 3 cases of calling MemSet appear to be unnecessary, as the zeroed out values are immediately overwritten. We need to zero out the holes in the structures so that hash functions work correctly. I suspect we are both wrong :-) These structures aren't used in any hashed structure that I can see, but the effect of the MemSet is to zero out the it_interval value of the itimer, making sure the timer is a "one shot" timer - so they are necessary after all, but for a different reason (it's been a while since I used setitimer, and the man page is less than a model of clarity). cheers andrew ---(end of broadcast)--- TIP 3: if posting/reading through Usenet, please send an appropriate subscribe-nomail command to [EMAIL PROTECTED] so that your message can get through to the mailing list cleanly
Re: [PATCHES] fix for strict-alias warnings
Tom Lane wrote: "Andrew Dunstan" <[EMAIL PROTECTED]> writes: Of course, the linux kernel is aimed at a limited set of compilers - as I understand it basically gcc although it has been made to build with Intel compilers icc once compiled the kernel. But they had to teach it quite a lots of gccisms. - which makes things somewhat easier for them. What is our target set of compilers? What is our target version of C? "Pretty much anything that speaks ANSI C" is my usual feeling about that. As yet we have not heard of any non-gcc compilers in which this is a problem, although you have a point that some compiler somewhere may do this and not have a way to turn it off :-( Intel's icc compiler supports strict alias analysis, but the default was off. Also note that uninhibited casting between types can still cause alignment problems, We understand that issue, we solved it years ago. BTW, I haven't looked at the problem spots in detail. How many of them are due to the use of MemSet in conjunction with other access to a chunk of memory? ISTM that we need not worry about code motion around a MemSet call, since that would require the compiler to prove that the memset() path through the macro wouldn't be affected, which I doubt it would think. gcc is quite good at propagating constants around. This is heavily used in the linux-kernel: __builtin_constant(x), and then large switch statements that are completely evaluated at compile time. There is a good chance that gcc figures out that MemSet(,0,sizeof(double)) are two writes to two integer values, and then decides that they can't alias with reads/write to the double. I'll search for a suitable gcc list and post the memset macro - that might give a definitive answer. -- Manfred ---(end of broadcast)--- TIP 3: if posting/reading through Usenet, please send an appropriate subscribe-nomail command to [EMAIL PROTECTED] so that your message can get through to the mailing list cleanly
Re: [PATCHES] fix for strict-alias warnings
Andrew Dunstan writes: > And looking at it even closer, 2 of the 3 cases of calling MemSet appear > to be unnecessary, as the zeroed out values are immediately overwritten. We need to zero out the holes in the structures so that hash functions work correctly. -- Peter Eisentraut [EMAIL PROTECTED] ---(end of broadcast)--- TIP 1: subscribe and unsubscribe commands go to [EMAIL PROTECTED]
Re: [PATCHES] fix for strict-alias warnings
Andrew Dunstan wrote: there were 3 calls to MemSet it complained about - all in src/backend/storage/lmgr/proc.c, and all zeroing out the timeval structure. (is MemSet actually a gain in this instance?) And looking at it even closer, 2 of the 3 cases of calling MemSet appear to be unnecessary, as the zeroed out values are immediately overwritten. cheers andrew ---(end of broadcast)--- TIP 3: if posting/reading through Usenet, please send an appropriate subscribe-nomail command to [EMAIL PROTECTED] so that your message can get through to the mailing list cleanly
Re: [PATCHES] fix for strict-alias warnings
Tom Lane wrote: BTW, I haven't looked at the problem spots in detail. How many of them are due to the use of MemSet in conjunction with other access to a chunk of memory? ISTM that we need not worry about code motion around a MemSet call, since that would require the compiler to prove that the memset() path through the macro wouldn't be affected, which I doubt it would think. there were 3 calls to MemSet it complained about - all in src/backend/storage/lmgr/proc.c, and all zeroing out the timeval structure. (is MemSet actually a gain in this instance?) there was the very odd one in src/bin/psql/command.c, which seems to me to be bogus there were 3 in src/backend/commands/tablecmds.c and src/backend/executor/execQual.c complaining about casting things to (Node *) finally, there was a warning about incompatible pointer types (i.e. not a type-pun warning) in the (uncast) call to shmdt in src/backend/port/sysv_shmem.c cheers andrew ---(end of broadcast)--- TIP 9: the planner will ignore your desire to choose an index scan if your joining column's datatypes do not match
Re: [PATCHES] fix for strict-alias warnings
"Andrew Dunstan" <[EMAIL PROTECTED]> writes: > Of course, the linux kernel is aimed at a limited set of compilers - as I > understand it basically gcc although it has been made to build with Intel > compilers - which makes things somewhat easier for them. What is our target > set of compilers? What is our target version of C? "Pretty much anything that speaks ANSI C" is my usual feeling about that. As yet we have not heard of any non-gcc compilers in which this is a problem, although you have a point that some compiler somewhere may do this and not have a way to turn it off :-( > Also note that uninhibited casting between types can still cause alignment > problems, We understand that issue, we solved it years ago. BTW, I haven't looked at the problem spots in detail. How many of them are due to the use of MemSet in conjunction with other access to a chunk of memory? ISTM that we need not worry about code motion around a MemSet call, since that would require the compiler to prove that the memset() path through the macro wouldn't be affected, which I doubt it would think. regards, tom lane ---(end of broadcast)--- TIP 3: if posting/reading through Usenet, please send an appropriate subscribe-nomail command to [EMAIL PROTECTED] so that your message can get through to the mailing list cleanly
Re: [PATCHES] fix for strict-alias warnings
- Original Message - From: "Tom Lane" <[EMAIL PROTECTED]> > > I have to agree now with Andrew's last mail that -fno-strict-aliasing is > the only safe solution. Since gcc isn't even pretending that it can > warn in all cases where the optimization might break things, I'm not > sure we could ever responsibly enable this optimization. I do not feel > this is our problem; it is the compiler hackers' fault if they need to > make unsupportable, untestable assumptions about application code. > > Basically, ISO broke the language here, and I say it's a screwup up with > which we need not put. > You and Linus Torvalds ;-) I recall seeing almost this exact discussion on the kernel hackers list a few years ago. Of course, the linux kernel is aimed at a limited set of compilers - as I understand it basically gcc although it has been made to build with Intel compilers - which makes things somewhat easier for them. What is our target set of compilers? What is our target version of C? (being unsure on these issues I gave my initdb.c a tour through "gcc -ansi -pedantic" at one stage). Also note that uninhibited casting between types can still cause alignment problems, quite apart from the strict aliasing issue (That was what confused Bruce, I think - the email on the netbsd list referred to both strict aliasing issues and misalignment issues). Still, that apparently hasn't been a problem up to now so we are probably OK on this one. cheers andrew ---(end of broadcast)--- TIP 2: you can get off all lists at once with the unregister command (send "unregister YourEmailAddressHere" to [EMAIL PROTECTED])
Re: [PATCHES] fix for strict-alias warnings
Bruce Momjian <[EMAIL PROTECTED]> writes: > Tom Lane wrote: >> I don't think we understand the dangers quite yet, and I think the >> patches applied to date constitute useless thrashing rather than fixes. > In reading http://www.gnu.org/software/gcc/bugs.html#nonbugs_c and the > link it references, > http://mail-index.netbsd.org/tech-kern/2003/08/11/0001.html, they seem > to be talking about any pointer casting. The latter reference makes it absolutely crystal-clear that inserting void* casts does *not* fix the issue. Also observe the statement > gcc may warn for some constructs that break the aliasing rules, but not > all of them (or not even most of them!), so a warning-free source code > does not give you any guarantee. I have to agree now with Andrew's last mail that -fno-strict-aliasing is the only safe solution. Since gcc isn't even pretending that it can warn in all cases where the optimization might break things, I'm not sure we could ever responsibly enable this optimization. I do not feel this is our problem; it is the compiler hackers' fault if they need to make unsupportable, untestable assumptions about application code. Basically, ISO broke the language here, and I say it's a screwup up with which we need not put. regards, tom lane ---(end of broadcast)--- TIP 9: the planner will ignore your desire to choose an index scan if your joining column's datatypes do not match
Re: [PATCHES] fix for strict-alias warnings
Tom Lane wrote: > "Andrew Dunstan" <[EMAIL PROTECTED]> writes: > > Even without the extra overhead, the danger of strict-aliasing is not just > > related to alignment. > > If I understand the issue at all, it has *nothing* to do with alignment. > > > As I understand it, given strict-aliasing assumptions > > the compiler is free to reorder some operations on things it thinks can't be > > the same thing, or even optimise them away because they can have no effect. > > Yah... > > > I'm not 100% sure we have avoided that danger. > > I don't think we understand the dangers quite yet, and I think the > patches applied to date constitute useless thrashing rather than fixes. > I'd like to see less quick-hack patching and more discussion. > > In particular, given that there is as yet no demonstrated effect other > than mere warnings issued by a possibly-buggy gcc release, I think it's > premature to be hacking our sources at all. OK, patch removed. When no one commented after 24 hours on my makeNode() idea, I though I was on to something. :-( In reading http://www.gnu.org/software/gcc/bugs.html#nonbugs_c and the link it references, http://mail-index.netbsd.org/tech-kern/2003/08/11/0001.html, they seem to be talking about any pointer casting. It also has this gem: I have seen some commits that "fix" gcc 3.3 alias warnings, that does not give me warm fuzzy feelings (the commits that is), and I have alse seen a lot of confusion about aliasing (and ISO C in general) on different mailing lists, so I have tried to explain some of the issues that I know have/will bite us. indicating they might remove these warnings soon anyway. I am not even going to point this gcc issue on the 7.4 open items list. -- Bruce Momjian| http://candle.pha.pa.us [EMAIL PROTECTED] | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup.| Newtown Square, Pennsylvania 19073 ---(end of broadcast)--- TIP 7: don't forget to increase your free space map settings
Re: [PATCHES] fix for strict-alias warnings
"Andrew Dunstan" <[EMAIL PROTECTED]> writes: > Even without the extra overhead, the danger of strict-aliasing is not just > related to alignment. If I understand the issue at all, it has *nothing* to do with alignment. > As I understand it, given strict-aliasing assumptions > the compiler is free to reorder some operations on things it thinks can't be > the same thing, or even optimise them away because they can have no effect. Yah... > I'm not 100% sure we have avoided that danger. I don't think we understand the dangers quite yet, and I think the patches applied to date constitute useless thrashing rather than fixes. I'd like to see less quick-hack patching and more discussion. In particular, given that there is as yet no demonstrated effect other than mere warnings issued by a possibly-buggy gcc release, I think it's premature to be hacking our sources at all. regards, tom lane ---(end of broadcast)--- TIP 7: don't forget to increase your free space map settings
Re: [PATCHES] fix for strict-alias warnings
- Original Message - From: "Tom Lane" <[EMAIL PROTECTED]> > Bruce Momjian <[EMAIL PROTECTED]> writes: > > I have attached and applied the following patch to use makeNode for > > structures that will later be cast to Node*, rather than having them be > > allocated as stack variables. > > AFAICT, this adds unnecessary palloc overhead without actually reducing > the risk of optimization problems. > Even without the extra overhead, the danger of strict-aliasing is not just related to alignment. As I understand it, given strict-aliasing assumptions the compiler is free to reorder some operations on things it thinks can't be the same thing, or even optimise them away because they can have no effect. I'm not 100% sure we have avoided that danger. Quote from nice page on this subject: "ISO C is not your grandfather's C". :-) cheers andrew ---(end of broadcast)--- TIP 7: don't forget to increase your free space map settings
Re: [PATCHES] fix for strict-alias warnings
Bruce Momjian <[EMAIL PROTECTED]> writes: > I have attached and applied the following patch to use makeNode for > structures that will later be cast to Node*, rather than having them be > allocated as stack variables. AFAICT, this adds unnecessary palloc overhead without actually reducing the risk of optimization problems. regards, tom lane ---(end of broadcast)--- TIP 1: subscribe and unsubscribe commands go to [EMAIL PROTECTED]
Re: [PATCHES] fix for strict-alias warnings
The warning is this:
command.c: In function `unescape':
command.c:1283: warning: dereferencing type-punned pointer will break
strict-aliasing rules
p is declared thus:
const unsigned char *p;
If I change common.c/h so that parse_char() takes an (unsigned char **)
argument, cast its 2nd argument to the call to strtol to (char **), and
change the cast in the call to (unsigned char **) those warnings go away,
but I get one from the other place parse_char() is called, namely prompt.c,
which has similar code but this time p is not declared as unsigned. - and so
it goes.
Grrr. What a pain.
andrew
- Original Message -
From: "Bruce Momjian" <[EMAIL PROTECTED]>
To: "Andrew Dunstan" <[EMAIL PROTECTED]>
Cc: "PG Patches" <[EMAIL PROTECTED]>
Sent: Sunday, October 12, 2003 7:18 PM
Subject: Re: [PATCHES] fix for strict-alias warnings
> Andrew Dunstan wrote:
> >
> > - Original Message -
> > From: "Bruce Momjian" <[EMAIL PROTECTED]>
> > >
> > > I have backed out the patch.
> > >
> > > Looking at the case in tablecmds.c and proc.c, the first was assigning
a
> > > struct with a NodeTag pointer as its first element to another struct
> > > with NodeTag as its first element. In fact, we do this all over the
> > > place, having different structure pointers with a start element of
> > > NodeTag.
>
> I have attached and applied the following patch to use makeNode for
> structures that will later be cast to Node*, rather than having them be
> allocated as stack variables.
>
> This leaves the only remaning compiler warning coming from common.c listed
> below. What is the exact warning generated --- this seems like a
> different issue.
>
> --
-
>
> Index: src/bin/psql/command.c
> ===
> RCS file: /projects/cvsroot/pgsql-server/src/bin/psql/command.c,v
> retrieving revision 1.103
> diff -c -w -r1.103 command.c
> *** src/bin/psql/command.c 29 Sep 2003 16:39:18 - 1.103
> --- src/bin/psql/command.c 11 Oct 2003 13:50:15 -
> ***
> *** 1280,1286
> case '7':
> case '8':
> case '9':
> ! c = parse_char((char **) &p);
> break;
>
> default:
> --- 1280,1286
> case '7':
> case '8':
> case '9':
> ! c = parse_char((void *) &p);
> break;
>
> default:
>
> --
> Bruce Momjian| http://candle.pha.pa.us
> [EMAIL PROTECTED] | (610) 359-1001
> + If your life is a hard drive, | 13 Roberts Road
> + Christ can be your backup.| Newtown Square, Pennsylvania
19073
>
> Index: src/backend/commands/tablecmds.c
> ===
> RCS file: /cvsroot/pgsql-server/src/backend/commands/tablecmds.c,v
> retrieving revision 1.88
> diff -c -c -r1.88 tablecmds.c
> *** src/backend/commands/tablecmds.c 11 Oct 2003 18:04:25 - 1.88
> --- src/backend/commands/tablecmds.c 12 Oct 2003 23:10:21 -
> ***
> *** 3449,3454
> --- 3449,3455
> Relation pkrel)
> {
> HeapScanDesc scan;
> + TriggerData *trigdata = makeNode(TriggerData); /* must be Node aligned
*/
> HeapTuple tuple;
> Trigger trig;
> List*list;
> ***
> *** 3506,3512
> while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
> {
> FunctionCallInfoData fcinfo;
> - TriggerData trigdata;
>
> /*
> * Make a call to the trigger function
> --- 3507,3512
> ***
> *** 3518,3537
> /*
> * We assume RI_FKey_check_ins won't look at flinfo...
> */
> ! trigdata.type = T_TriggerData;
> ! trigdata.tg_event = TRIGGER_EVENT_INSERT | TRIGGER_EVENT_ROW;
> ! trigdata.tg_relation = rel;
> ! trigdata.tg_trigtuple = tuple;
> ! trigdata.tg_newtuple = NULL;
> ! trigdata.tg_trigger = &trig;
>
> ! fcinfo.context = (Node *) &trigdata;
>
> RI_FKey_check_ins(&fcinfo);
> }
>
> heap_endscan(scan);
>
> pfree(trig.tgargs);
> }
>
> --- 3518,3538
> /*
> * We assume RI_FKey_check_ins won't look at flinfo...
> */
> ! trigdata->type = T_TriggerData;
> ! trigdata->tg_event = TRIGGER_EVENT_INSERT | TRIGGER_EVENT_ROW;
> ! trigdata->tg_relation = rel;
> ! trigdata->tg_trigtuple = tuple;
> ! trigdata->tg_newtuple = NULL;
> ! trigdata->tg_trigger = &trig;
>
&
Re: [PATCHES] fix for strict-alias warnings
Bruce Momjian wrote: > Andrew Dunstan wrote: > > > > - Original Message - > > From: "Bruce Momjian" <[EMAIL PROTECTED]> > > > > > > I have backed out the patch. > > > > > > Looking at the case in tablecmds.c and proc.c, the first was assigning a > > > struct with a NodeTag pointer as its first element to another struct > > > with NodeTag as its first element. In fact, we do this all over the > > > place, having different structure pointers with a start element of > > > NodeTag. > > I have attached and applied the following patch to use makeNode for > structures that will later be cast to Node*, rather than having them be > allocated as stack variables. Oh, and thanks to everyone for doing the research on this compiler issue. -- Bruce Momjian| http://candle.pha.pa.us [EMAIL PROTECTED] | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup.| Newtown Square, Pennsylvania 19073 ---(end of broadcast)--- TIP 2: you can get off all lists at once with the unregister command (send "unregister YourEmailAddressHere" to [EMAIL PROTECTED])
Re: [PATCHES] fix for strict-alias warnings
Andrew Dunstan wrote:
>
> - Original Message -
> From: "Bruce Momjian" <[EMAIL PROTECTED]>
> >
> > I have backed out the patch.
> >
> > Looking at the case in tablecmds.c and proc.c, the first was assigning a
> > struct with a NodeTag pointer as its first element to another struct
> > with NodeTag as its first element. In fact, we do this all over the
> > place, having different structure pointers with a start element of
> > NodeTag.
I have attached and applied the following patch to use makeNode for
structures that will later be cast to Node*, rather than having them be
allocated as stack variables.
This leaves the only remaning compiler warning coming from common.c listed
below. What is the exact warning generated --- this seems like a
different issue.
---
Index: src/bin/psql/command.c
===
RCS file: /projects/cvsroot/pgsql-server/src/bin/psql/command.c,v
retrieving revision 1.103
diff -c -w -r1.103 command.c
*** src/bin/psql/command.c 29 Sep 2003 16:39:18 - 1.103
--- src/bin/psql/command.c 11 Oct 2003 13:50:15 -
***
*** 1280,1286
case '7':
case '8':
case '9':
! c = parse_char((char **) &p);
break;
default:
--- 1280,1286
case '7':
case '8':
case '9':
! c = parse_char((void *) &p);
break;
default:
--
Bruce Momjian| http://candle.pha.pa.us
[EMAIL PROTECTED] | (610) 359-1001
+ If your life is a hard drive, | 13 Roberts Road
+ Christ can be your backup.| Newtown Square, Pennsylvania 19073
Index: src/backend/commands/tablecmds.c
===
RCS file: /cvsroot/pgsql-server/src/backend/commands/tablecmds.c,v
retrieving revision 1.88
diff -c -c -r1.88 tablecmds.c
*** src/backend/commands/tablecmds.c11 Oct 2003 18:04:25 - 1.88
--- src/backend/commands/tablecmds.c12 Oct 2003 23:10:21 -
***
*** 3449,3454
--- 3449,3455
Relation pkrel)
{
HeapScanDesc scan;
+ TriggerData *trigdata = makeNode(TriggerData); /* must be Node aligned */
HeapTuple tuple;
Trigger trig;
List *list;
***
*** 3506,3512
while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
{
FunctionCallInfoData fcinfo;
- TriggerData trigdata;
/*
* Make a call to the trigger function
--- 3507,3512
***
*** 3518,3537
/*
* We assume RI_FKey_check_ins won't look at flinfo...
*/
! trigdata.type = T_TriggerData;
! trigdata.tg_event = TRIGGER_EVENT_INSERT | TRIGGER_EVENT_ROW;
! trigdata.tg_relation = rel;
! trigdata.tg_trigtuple = tuple;
! trigdata.tg_newtuple = NULL;
! trigdata.tg_trigger = &trig;
! fcinfo.context = (Node *) &trigdata;
RI_FKey_check_ins(&fcinfo);
}
heap_endscan(scan);
pfree(trig.tgargs);
}
--- 3518,3538
/*
* We assume RI_FKey_check_ins won't look at flinfo...
*/
! trigdata->type = T_TriggerData;
! trigdata->tg_event = TRIGGER_EVENT_INSERT | TRIGGER_EVENT_ROW;
! trigdata->tg_relation = rel;
! trigdata->tg_trigtuple = tuple;
! trigdata->tg_newtuple = NULL;
! trigdata->tg_trigger = &trig;
! fcinfo.context = (Node *) trigdata;
RI_FKey_check_ins(&fcinfo);
}
heap_endscan(scan);
+ pfree(trigdata);
pfree(trig.tgargs);
}
Index: src/backend/executor/execQual.c
===
RCS file: /cvsroot/pgsql-server/src/backend/executor/execQual.c,v
retrieving revision 1.148
diff -c -c -r1.148 execQual.c
*** src/backend/executor/execQual.c 11 Oct 2003 18:04:25 - 1.148
--- src/backend/executor/execQual.c 12 Oct 2003 23:10:23 -
***
*** 699,705
List *arguments = fcache->args;
Datum result;
FunctionCallInfoData fcinfo;
! ReturnSetInfo rsinfo; /* for functions returning sets */
ExprDoneCond argDone;
Re: [PATCHES] fix for strict-alias warnings
Andrew Dunstan wrote:
> > The proc.c cases were using MemSet, which was checking if the
> > int* as aligned for int* access. In fact, we could change MemSet to
> > always take a void *, and do the int* casting when we access it after
> > testing for alignment.
> >
>
> Since MemSet is generic, that is probably a good idea.
I have applied the following patch to add an additional void* cast to
MemSet, and added a comment that this was safe because alignment is
checked below --- patch attached.
--
Bruce Momjian| http://candle.pha.pa.us
[EMAIL PROTECTED] | (610) 359-1001
+ If your life is a hard drive, | 13 Roberts Road
+ Christ can be your backup.| Newtown Square, Pennsylvania 19073
Index: src/include/c.h
===
RCS file: /cvsroot/pgsql-server/src/include/c.h,v
retrieving revision 1.153
diff -c -c -r1.153 c.h
*** src/include/c.h 21 Sep 2003 17:57:21 - 1.153
--- src/include/c.h 11 Oct 2003 19:51:09 -
***
*** 604,610
#define MemSet(start, val, len) \
do \
{ \
! int32 * _start = (int32 *) (start); \
int _val = (val); \
Size_len = (len); \
\
--- 604,611
#define MemSet(start, val, len) \
do \
{ \
! /* (void *) used because we check for alignment below */ \
! int32 * _start = (int32 *) (void *) (start); \
int _val = (val); \
Size_len = (len); \
\
---(end of broadcast)---
TIP 9: the planner will ignore your desire to choose an index scan if your
joining column's datatypes do not match
Re: [PATCHES] fix for strict-alias warnings
- Original Message -
From: "Bruce Momjian" <[EMAIL PROTECTED]>
>
> I have backed out the patch.
>
> Looking at the case in tablecmds.c and proc.c, the first was assigning a
> struct with a NodeTag pointer as its first element to another struct
> with NodeTag as its first element. In fact, we do this all over the
> place, having different structure pointers with a start element of
> NodeTag.
Right - and it is what would have to change if you really want to obey the
ISO C rules, I believe. This is handled easily in other languages using
variant records, but C is kinda primitive here :-)
As I understand it, instead of
struct foo {
int tag
foostuff f;
}
struct bar {
int tag;
barstuff b;
}
you would need to do something like
struct foo {
foostuff f;
};
struct bar {
barstuff b;
};
struct foobar {
int tag;
union {
struct foo foo;
struct bar bar;
} v;
};
> The proc.c cases were using MemSet, which was checking if the
> int* as aligned for int* access. In fact, we could change MemSet to
> always take a void *, and do the int* casting when we access it after
> testing for alignment.
>
Since MemSet is generic, that is probably a good idea.
> The big question in my mind is whether there there is other struct *
> passing that could be masked right now by void* casting, and if so, do
> they have different first elements? This determined whether we do
> -fstrict-aliasing for gcc, or fix just these few cases.
Just analysing this is a non-trivial piece of work, I think.
cheers
andrew
---(end of broadcast)---
TIP 8: explain analyze is your friend
Re: [PATCHES] fix for strict-alias warnings
Andrew Dunstan wrote: > Tough words! :-) > > ISTM the best thing would be to back out the patch, add -fno-strict-aliasing > for gcc, and add a TODO to fix this thoroughly. > > Having -fstrict-aliasing on and ignoring the warnings doesn't seem like a > sound strategy. I think we should fix it or turn it off. The web is littered > with projects that got bizzare happenings when they turned it on without any > accompanying code changes. > > I agree with Tom that my patch isn't ideal (I thought I said as much). > Fixing it thoroughly will require some significant code changes, though. We > seem to be far too close to 7.4 release to contemplate that. I have backed out the patch. Looking at the case in tablecmds.c and proc.c, the first was assigning a struct with a NodeTag pointer as its first element to another struct with NodeTag as its first element. In fact, we do this all over the place, having different structure pointers with a start element of NodeTag. The proc.c cases were using MemSet, which was checking if the int* as aligned for int* access. In fact, we could change MemSet to always take a void *, and do the int* casting when we access it after testing for alignment. The big question in my mind is whether there there is other struct * passing that could be masked right now by void* casting, and if so, do they have different first elements? This determined whether we do -fstrict-aliasing for gcc, or fix just these few cases. -- Bruce Momjian| http://candle.pha.pa.us [EMAIL PROTECTED] | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup.| Newtown Square, Pennsylvania 19073 ---(end of broadcast)--- TIP 5: Have you checked our extensive FAQ? http://www.postgresql.org/docs/faqs/FAQ.html
Re: [PATCHES] fix for strict-alias warnings
Tough words! :-) ISTM the best thing would be to back out the patch, add -fno-strict-aliasing for gcc, and add a TODO to fix this thoroughly. Having -fstrict-aliasing on and ignoring the warnings doesn't seem like a sound strategy. I think we should fix it or turn it off. The web is littered with projects that got bizzare happenings when they turned it on without any accompanying code changes. I agree with Tom that my patch isn't ideal (I thought I said as much). Fixing it thoroughly will require some significant code changes, though. We seem to be far too close to 7.4 release to contemplate that. cheers andrew - Original Message - From: "Tom Lane" <[EMAIL PROTECTED]> To: "Bruce Momjian" <[EMAIL PROTECTED]> Cc: "Andrew Dunstan" <[EMAIL PROTECTED]>; "PG Patches" <[EMAIL PROTECTED]> Sent: Saturday, October 11, 2003 1:29 PM Subject: Re: [PATCHES] fix for strict-alias warnings > Bruce Momjian <[EMAIL PROTECTED]> writes: > > I don't have a version that does the double-cast, but I still have the > > patch to back out and put in a new one. Andrew's point was that we cast > > to void * in many places, so this case is not unique. Is that wrong? > > I do not like code that uses cast to void* as a substitute for casting > to the real destination type. I think it's a lazy substitute for > providing the correct cast, and it renders the code more fragile because > there is *no* possibility of the compiler detecting a problem should you > change the source or destination datatype in a way that renders the cast > wrong. > > I have not gone around and tried to fix all the places that are lazy in > this way, but I don't want to introduce more, and for sure I don't want > to set a precedent that we'll weaken our type checking any time gcc > burps for ill-defined reasons. > > I agree completely with all of the objections you raised in your > original comment on the patch. In particular, I don't think we > understand why gcc is complaining about these few places and not any of > the thousands of other casts in our code. Until we understand that > difference completely, we are not "fixing a bug" by introducing void* > casts. I'd have to call it cargo-cult programming instead. > > I am perfectly content to leave the warnings in place until we have a > satisfactory explanation. > > regards, tom lane > > ---(end of broadcast)--- > TIP 2: you can get off all lists at once with the unregister command > (send "unregister YourEmailAddressHere" to [EMAIL PROTECTED]) ---(end of broadcast)--- TIP 8: explain analyze is your friend
Re: [PATCHES] fix for strict-alias warnings
Bruce Momjian <[EMAIL PROTECTED]> writes: > I don't have a version that does the double-cast, but I still have the > patch to back out and put in a new one. Andrew's point was that we cast > to void * in many places, so this case is not unique. Is that wrong? I do not like code that uses cast to void* as a substitute for casting to the real destination type. I think it's a lazy substitute for providing the correct cast, and it renders the code more fragile because there is *no* possibility of the compiler detecting a problem should you change the source or destination datatype in a way that renders the cast wrong. I have not gone around and tried to fix all the places that are lazy in this way, but I don't want to introduce more, and for sure I don't want to set a precedent that we'll weaken our type checking any time gcc burps for ill-defined reasons. I agree completely with all of the objections you raised in your original comment on the patch. In particular, I don't think we understand why gcc is complaining about these few places and not any of the thousands of other casts in our code. Until we understand that difference completely, we are not "fixing a bug" by introducing void* casts. I'd have to call it cargo-cult programming instead. I am perfectly content to leave the warnings in place until we have a satisfactory explanation. regards, tom lane ---(end of broadcast)--- TIP 2: you can get off all lists at once with the unregister command (send "unregister YourEmailAddressHere" to [EMAIL PROTECTED])
Re: [PATCHES] fix for strict-alias warnings
Tom Lane wrote: > Bruce Momjian <[EMAIL PROTECTED]> writes: > > Patch applied. Thanks. > > I hope you applied it with the additional changes you asked for --- > at the very least, cast to (void*) and then to the destination type. > As-is, the patch simply suppresses all error detection for these > conversions, which seems a bad move. I don't have a version that does the double-cast, but I still have the patch to back out and put in a new one. Andrew's point was that we cast to void * in many places, so this case is not unique. Is that wrong? -- Bruce Momjian| http://candle.pha.pa.us [EMAIL PROTECTED] | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup.| Newtown Square, Pennsylvania 19073 ---(end of broadcast)--- TIP 2: you can get off all lists at once with the unregister command (send "unregister YourEmailAddressHere" to [EMAIL PROTECTED])
Re: [PATCHES] fix for strict-alias warnings
Bruce Momjian <[EMAIL PROTECTED]> writes: > Patch applied. Thanks. I hope you applied it with the additional changes you asked for --- at the very least, cast to (void*) and then to the destination type. As-is, the patch simply suppresses all error detection for these conversions, which seems a bad move. regards, tom lane ---(end of broadcast)--- TIP 6: Have you searched our list archives? http://archives.postgresql.org
Re: [PATCHES] fix for strict-alias warnings
Patch applied. Thanks. --- Andrew Dunstan wrote: > > This patch will stop gcc from issuing warnings about type-punned objects when > -fstrict-aliasing is turned on, as it is in the latest gcc when you use -O2 > > enjoy > > andrew > [ Attachment, skipping... ] > > ---(end of broadcast)--- > TIP 4: Don't 'kill -9' the postmaster -- Bruce Momjian| http://candle.pha.pa.us [EMAIL PROTECTED] | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup.| Newtown Square, Pennsylvania 19073 ---(end of broadcast)--- TIP 2: you can get off all lists at once with the unregister command (send "unregister YourEmailAddressHere" to [EMAIL PROTECTED])
