source/Interpreter/Makefile: broken install-local

[llvmbot]

Bugzilla Link	10157
Resolution	FIXED
Resolved on	Jun 23, 2011 14:06
Version	unspecified
OS	All
Attachments	don't depend on an installed lib; proper symlink, fix PYTHON_DEST_DIR as well
Reporter	LLVM Bugzilla Contributor

Extended Description

I think the attached patch is in order.

[llvmbot]

One comment on your patch. Why does the path to the LLDB library not respect the DESTDIR variable? liblldb will be installed to $(DESTDIR)$(PROJ_libdir), but the symlink is to $(PROJ_libdir)/liblldb.so. Is this on purpose? I could see how, if e.g. the DESTDIR were the root of a chroot(8) environment. But, are you sure you want to do it this way?

[llvmbot]

Yes, that's intentional.

[llvmbot]

use realpath as well
What this patch does:

Previously, PYTHON_DEST_DIR did not respect DESTDIR contrary to what one might expect from the name. it now does. Before prepending $(DESTDIR), make's realpath function is used. On a system like mine,

python -c 'import sys; print sys.exec_prefix'

returns

/usr/bin/../../etc/env.d/alternatives/python2/_current/usr/bin/../../../../../../../usr

which is equivalent to /usr. But since ${DESTDIR}/etc usually does not exist, preprending $(DESTDIR) without calling realpath first leads to a broken path.

Previously, install-local depended on $(LIBLLDB) which was $(DESTDIR)$(...). Calling make install-local would fail because there was no rule to build that.

Furthermore, the a symlink pointed to $(LIBLLDB), thus to something inside $(DESTDIR) which nothing ever should (it simply does not make sense).

[llvmbot]

Almost. You have to $(call realpath), because if GNU make doesn't provide it, the LLVM makefile system defines its own realpath function. Also, you need to regenerate your patch against HEAD, because Peter Collingbourne committed his series of patches, two of which change source/Interpreter/Makefile.

[llvmbot]

patch against current trunk
Like this?

[llvmbot]

Great! I committed your patch in r133689. Thanks for the patch.

Oh, and in the future, you should send your LLDB patches to lldb-commits.

[llvmbot]

Oops...

$(call realpath [..]) should've been $(call realpath,[..]) (note the comma) I guess.

[llvmbot]

Yep you're right, the arguments to realpath should have been separated from the name 'realpath' by commas. I fixed it in r133744. Thanks!