[enhancement] set install_name on Darwin platforms

[grobian]

Below is a "trivial" patch :)

Makefile: set install_name on Darwin

https://bugs.gentoo.org/show_bug.cgi?id=546098

--- a/Makefile
+++ b/Makefile
@@ -62,6 +62,8 @@
 ifneq (darwin,$(PLATFORM))
 # TODO(bnoordhuis) The native SunOS linker expects -h rather than -soname...
 LDFLAGS_LIB += -Wl,-soname=$(SONAME)
+else
+LDFLAGS_LIB += -Wl,-install_name,$(LIBDIR)/$(SONAME)
 endif

 test: test_g test_fast

Note that while I use $(SONAME) for consistency, it feels odd because it includes the full version number. SONAME/install_name typically is something like libhttp_parser.2.dylib (or libhttp_parser.so.2), referencing the major version component only. But from the looks of it, this seems intentional, suggesting e.g. 2.6 and 2.7 release are ABI incompatible, as well as 2.6.2 and 2.6.3 would. (E.g. currently any http-parser update will break all consumers.)

[bnoordhuis]

I think a pull request would be welcome. I'd flip the clauses and use ifeq (darwin,...) for clarity.

Note that while I use $(SONAME) for consistency, it feels odd because it includes the full version number. [..] suggesting e.g. 2.6 and 2.7 release are ABI incompatible, as well as 2.6.2 and 2.6.3 would.

When I added the soname for #151, I included the minor because minor releases are not backwards compatible with older minor releases (they add new APIs.) The patch number is not supposed to be in there, though.

cc @indutny since you landed 2896229 which added that.

[grobian]

It was introduced in 36f107f IMO (perhaps using $(MAJ) and $(MIN) helps to avoid this issue from re-appearing)

[grobian]

see PR #358 I left the versioning bits aside, but can do a suggestion for that too if you like.

[bnoordhuis]

You mean the MAJOR/MINOR thing? Sounds good.

[grobian]

as you wish ;)