| 1 |
Well, the problem is solved and the solution may be of interest |
| 2 |
to some Gentoo users. |
| 3 |
|
| 4 |
At first I thought that I was in for some heavy debugging, but the |
| 5 |
fault was a simple matter (simple at least superficially). |
| 6 |
|
| 7 |
I have been using the latest gcc-4.5.1 for some time now and |
| 8 |
I try to compile my system with all the newest optimizations, namely |
| 9 |
the "-flto," "-floop-interchange," "-floop-strip-mine," and |
| 10 |
"-floop-block" compiler flags which enables Graphite and LTO. |
| 11 |
Usually I have no problems during an emerge but when I do the |
| 12 |
temporary removal of the "-flto" option allows the emerge to finish. |
| 13 |
So far there have been no issues with these flags. |
| 14 |
|
| 15 |
Now, whenever a problem with software is encountered the first question |
| 16 |
to ask is what has been changed on the system. In my case it most |
| 17 |
likely was the use of these compiler flags. So, on a lark, I decided |
| 18 |
to re-emerge ghostscript with a basic set of CFLAGS = -O2 -march=native. |
| 19 |
It worked. The problem I was having with all black images has |
| 20 |
disappeared and ghostscript functions normally. |
| 21 |
|
| 22 |
However, I really cannot consider this to be normal behavior. Compiler |
| 23 |
optimization should not degrade the function of a program. Something |
| 24 |
subtle must be at fault with the gcc compiler. But being quite uninformed |
| 25 |
about compiler design I cannot say much more. |
| 26 |
|
| 27 |
Anyway, thanks again for all those who did the test. This confirmed |
| 28 |
that the problem was not ghostscript per se but rather a strange effect |
| 29 |
on the ghostscript code caused by the compiler optimization. |
| 30 |
|
| 31 |
I can now see why the Gentoo deveolpers are being so cautious by considering |
| 32 |
these latest GCC options as being experimental. |
| 33 |
|
| 34 |
Frank Peters |
Archives