Re: Continued: egcs-971201: testsuite results for i586-pc-linux-gnulibc1

Re: Continued: egcs-971201: testsuite results for i586-pc-linux-gnulibc1

[Jeffrey A Law]

  In message < u9pvnfb2kr.fsf@yorick.cygnus.com >you write:
  > Jeff, want to add something to the release notes to the effect that
  > -Wreturn-type and -pedantic cause template-heavy code to suffer from
  > compiler explosion?
It's already in the online faq under virtual memory exhaused.

Jeff

Re: Continued: egcs-971201: testsuite results for i586-pc-linux-gnulibc1

[Jason Merrill]

>>>>> Max Lawson <mlawson@drfmc.ceng.cea.fr> writes:

> $g++ -Wall -O test.cc 
> test.cc: In function `int main()':
> test.cc:10: virtual memory exhausted

As I said before, you need to add -Wno-return-type or the old memory
explosion bug resurfaces.  This is a known bug.  The problem is that
save_for_inline_copying wastes a ridiculous amount of memory in the
presence of recursive function definitions, which you get with templates.
With -Wno-return-type, we don't need to use save_for_inline_copying.

Jeff, want to add something to the release notes to the effect that
-Wreturn-type and -pedantic cause template-heavy code to suffer from
compiler explosion?

Jason

Re: Continued: egcs-971201: testsuite results for i586-pc-linux-gnulibc1

[Jason Merrill]

>>>>> Max Lawson <mlawson@drfmc.ceng.cea.fr> writes:

> I do not need to use any warning flag to obtain memory explosion. The 
> -Wno-return-type flag indeed solve the posted testcase problem, but can't 
> help. 
> I am now aware that my problem is mentioned in the first part of the online 
> faq (subject: virtual memory error). But what's called a large file ? And 
> how to simplify files ? In order to use template one has to merge header and 
> source files .... 

You can simplify by instantiating some templates in one file, and some in
another.  It's hard to answer general questions like this.

Jason

Re: Continued: egcs-971201: testsuite results for i586-pc-linux-gnulibc1

[Max Lawson]

> From jason@cygnus.com Wed Dec  3 03:09 MET 1997
> 
> >>>>> Max Lawson <mlawson@drfmc.ceng.cea.fr> writes:
> 
> > $g++ -Wall -O test.cc 
> > test.cc: In function `int main()':
> > test.cc:10: virtual memory exhausted
> 
> As I said before, you need to add -Wno-return-type or the old memory
> explosion bug resurfaces.  This is a known bug.  The problem is that
> save_for_inline_copying wastes a ridiculous amount of memory in the
> presence of recursive function definitions, which you get with templates.
> With -Wno-return-type, we don't need to use save_for_inline_copying.


I do not need to use any warning flag to obtain memory explosion. The 
-Wno-return-type flag indeed solve the posted testcase problem, but can't 
help. 
I am now aware that my problem is mentioned in the first part of the online 
faq (subject: virtual memory error). But what's called a large file ? And 
how to simplify files ? In order to use template one has to merge header and 
source files .... 


Thanx for the answers, Max.

Continued: egcs-971201: testsuite results for i586-pc-linux-gnulibc1

[Max Lawson]

	..  the typename5.C failure has left :-) 

Nevertheless, programs I have that are heavily relying on templates still 
make the system run out of memory depending on the optimization level. I'm 
trying to narrow down the problem to a simple testcase, but :^( .. 

Anyway, this is a _severe_ limitation for large programs :-(. or even 
small programs... {:-((

For example, the following testcase quoted from "egcs-bugs/1997-Nov/0235.html" 
still crash at compile time with the current 971201-snapshot on my box:

// --- test.cc ---
#include <iostream.h>
#include <list>
#include <algorithm>
using namespace std;

int main()
{
    int daten [16] = { 1, 4, 4, 6, 1, 2, 2, 3, 6, 6, 6, 5, 7, 5, 4, 4};
    list<int> menge;
    copy (daten, daten+16, back_inserter(menge));

    return 0;
}
// --- cc.tset ---

$g++ -Wall -O test.cc 
test.cc: In function `int main()':
test.cc:10: virtual memory exhausted

This happens in the call to copy; this let's me wonder if this is not 
related to non-trivial template instantiations of functions arguments. 
That's my 0.00000000000000000000000000000000000000000002cts guess.

(Note that I use to (I have to) call the compiler with no warning flags 
for my programs, therefore there is in my case no warning/optimization flags 
combinations.) 



Anyhow, thanx a lot for the job !-) 

Best Regards, Max.


----------
libc.5.39 & ld.so.1.9.5 & binutils-2.8.1.0.15
./configure --enable-shared
BOOT_CFLAGS="-O6 -g"

----------

Test Run By lawson on Tue Dec  2 13:45:33 1997
Native configuration is i586-pc-linux-gnulibc1

		=== gcc Summary ===

# of expected passes		4883
# of expected failures		5
# of unsupported tests		7
/tmp/egcs-971201/gcc/xgcc version egcs-2.90.20 971201 (gcc2-970802 experimental)

		=== g++ tests ===

XPASS: g++.jason/destruct3.C - (test for bogus messages, line 38)
XPASS: g++.mike/dyncast1.C  Execution test
XPASS: g++.mike/dyncast2.C  Execution test

		=== g++ Summary ===

# of expected passes		3400
# of unexpected successes	3
# of expected failures		80
# of untested testcases		6
/tmp/egcs-971201/gcc/testsuite/../xgcc version egcs-2.90.20 971201 (gcc2-970802 experimental)


		=== libio Summary ===

# of expected passes		40


		=== libstdc++ Summary ===

# of expected passes		30