log4cxx @0.10.0: Undefined local symbol L318

[peterhansgoldstein@…]

OS 10.6.8, XCode 3.2.6

Build fails, kicking out "Undefined local symbol L318"

Log file attached, but it's highly inscrutable to me.

[peterhansgoldstein@…]

Log file for this build attempt

[ryandesign (Ryan Carsten Schmidt)]

I don't believe we have a port called "ros-log4cxx" in the official MacPorts repository... Did you perhaps get this Portfile from somewhere else?

[ryandesign (Ryan Carsten Schmidt)]

Replying to ryandesign@…:

Did you perhaps get this Portfile from somewhere else?

Maybe from ​http://ros.org/ a few years ago?

[ryandesign (Ryan Carsten Schmidt)]

Or in fact recently. I've now found the ​ROS OS X MacPorts installation page, the ​ros-macports github repository, and the ​ros-log4cxx port inside it. This is not our bug, because this is not our port. Please report the problem to the ROS people.

[ryandesign (Ryan Carsten Schmidt)]

As it happens, I see the same problem with the log4cxx port that we do have. It happens on our Snow Leopard buildbot only, not on the newer buildbots. The Snow Leopard buildbot is the only one that uses gcc-4.2 as its default compiler, and I'm able to reproduce the problem on Mavericks if I use configure.compiler=apple-gcc-4.2 so it's some problem with old GCC. After some trial and error, I found that switching the optimization flags from -Os (the default as of MacPorts 2.2) back to -O2 (the default in previous versions of MacPorts) fixes it, so that seems the simplest fix.

[ryandesign (Ryan Carsten Schmidt)]

proposed patch

[ryandesign (Ryan Carsten Schmidt)]

There are no occurrences of the string "L318" in the source code. My thought is that perhaps this is a compiler-generated symbol name; the name makes me think it might be this file's 318th compiler-generated label, such as might be needed for a switch statement, of which there are many in the last file mentioned in the log, properties.cpp. Perhaps there is something unusual about the switch statements in properties.cpp that makes gcc-4.2's optimizer unhappy; some qualities that come to mind are that there are multiple nested switch statements, and that one of them has case statements after the default case (usually the default case is the last one).

I did find an old ​gcc bug when compiling a large switch statement; the error message is different, but it wasn't solved until gcc 4.3.

[mf2k (Frank Schima)]

Port abandoned. See #43811.

[Ken <21211439+kencu@…>]

In 84b96c5d5982d60682aaac58319d1c0ef4a29c59/macports-ports (master):

log4cxx: blacklist gcc-4.2 on darwin 10

closes: #40090