Summary: | configure should check gcj version | ||
---|---|---|---|
Product: | classpath | Reporter: | Nick <nickols_k> |
Component: | classpath | Assignee: | Dalibor Topic <robilad> |
Status: | RESOLVED FIXED | ||
Severity: | normal | CC: | bug-classpath |
Priority: | P3 | ||
Version: | 0.93 | ||
Target Milestone: | 0.97 | ||
Host: | i686-pc-linux-gnu | Target: | i686-pc-linux-gnu |
Build: | i686-pc-linux-gnu | Known to work: | |
Known to fail: | Last reconfirmed: | 2007-07-31 16:23:41 |
Description
Nick
2007-01-05 16:21:00 UTC
Which version of gcj are you using to compile? Note that the INSTALL notes say GCJ 4.0+ (part of the GNU GCC package), but in reality you might actually need 4.1+. I'm using gcc-3.4.6! Btw, it would be better to test version of GCJ during configuring phase!!! I tried to compile with gcc-4.1.1 but result is the same!!! Yeah, you are right, the configure script should have checked for that. Are you sure you configured with the new gcj in the PATH? Is the build really using the new gcj? For gcc-4.1.1 I'm running this command: CC=gcc4 ./configure And which gcj does configure detect? You might need/want to use --with-gcj=/path/to/gcj Thanks! With using the key: --with-gcj=/usr/local/bin/gcj I was able to compile classpath-0.93! /usr/local/bin/gcj --version gcj (GCC) 4.1.1 Thanks for trying and sorry that configure didn't detect the right gcj to use. I'll keep this bug open, but retitled it. For 0.94 we need ecj or gcj that understands the 1.5 language extensions. We need to add a configure check for that. I think we could use http://autoconf-archive.cryp.to/ac_try_compile_java.html to compile a small test file using generics, and abort configure if that fails. We do this now by checking that the compiler (whether ecj, javac or gcj) can compile 1.5 code. Do we still need a check for gcc however? I don't think we need a specific gcc check, as distributions tend to ship gcj in mashed up versions that don't really correspond to a specific FSF release, so testing for a gcc version wouldn't be nearly as useful as the current test. I believe this is now fixed. Closing. |