gcc48 @4.8.4_0, gcc49 @4.9.2_1: bootstrap comparison fails with clang-602.0.49 from Xcode 6.3

[howarth.at.macports@…]

The new Xcode 6.3's clang 6.1 (3.6.0svn) compilers break the bootstrap of FSF gcc 4.9.2 and gcc 4.9.3svn due to configure leaving BUILD_CONFIG unset.

​https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65733
​https://llvm.org/bugs/show_bug.cgi?id=23192

This same issue occurs when the llvm 3.5.1, 3.5.2 or 3.6.0 compilers are used to bootstrap gcc. The current workaround is to explicitly pass --with-build-config=bootstrap-debug to configure when the newer clang compilers are in use.

[ryandesign (Ryan Carsten Schmidt)]

Note that the Cc field requires complete email addresses.

[howarth.at.macports@…]

fix for ​https://llvm.org/bugs/show_bug.cgi?id=22046

[howarth.at.macports@…]

Note that if you plan to support building FSF gcc with the llvm.org clang compilers in MacPorts the proposed patch needs to be extended to capture clang 3.5 and later.

[ryandesign (Ryan Carsten Schmidt)]

Has duplicate #47419.

[ryandesign (Ryan Carsten Schmidt)]

Has duplicate #47417.

[howarth.at.macports@…]

FYI, the cause of the bug is that the described at ​http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20141222/250134.html.

[howarth.at.macports@…]

Just to be clear, this is considered a defect in both llvm.org and Apple clang so no hack will be forthcoming from the FSF gcc developers. Upstream lllvm.org needs to eliminate this regression in the improper setting of the CIE version in the absence of debug flags.

[sears3@…]

So basically no gcc version can be compiled until it's handled upstream? Effectively, new installs of macports are not happening for a while then (as I just figured out). Anybody have a hack/workaround?

[sears3@…]

Cc Me!

[larryv (Lawrence Velázquez)]

Replying to howarth.at.macports@…:

Note that if you plan to support building FSF gcc with the llvm.org clang compilers in MacPorts the proposed patch needs to be extended to capture clang 3.5 and later.

Is there a reason we can’t apply the workaround universally?

[howarth.at.macports@…]

Replying to sears3@…:

So basically no gcc version can be compiled until it's handled upstream? Effectively, new installs of macports are not happening for a while then (as I just figured out). Anybody have a hack/workaround?

No. You just have to explicitly pass '--with-build-config=bootstrap-debug' to configure when building with the clang 3.5 or later from llvm.org or with Apple clang 6.1. The code in configure that tests for proper debug code in object files is tripped up by this breakage in clang (where the wrong CIE version of dwarf 3 is emitted in the absence of debug flags).

[howarth.at.macports@…]

Replying to larryv@…:

Replying to howarth.at.macports@…:

Note that if you plan to support building FSF gcc with the llvm.org clang compilers in MacPorts the proposed patch needs to be extended to capture clang 3.5 and later.

Is there a reason we can’t apply the workaround universally?

None other than it is completely unnecessary for the earlier clang compilers (and hopefully for later clang compilers once the bug is fixed). Gcc upstream is not going to work around the issue and expects llvm to fix their compilers.

[larryv (Lawrence Velázquez)]

Taking over due to maintainer timeout.

[larryv (Lawrence Velázquez)]

Corroborated ​a comment on GCC PR bootstrap/65733 that gcc48 is also affected. gcc47 @4.7.4_3 and gcc5 @5-20150405 are not.

[larryv (Lawrence Velázquez)]

r135110