This is the mail archive of the
binutils@sourceware.org
mailing list for the binutils project.
[PATCH 2/2] LD/testsuite: Expand STB_GNU_UNIQUE test coverage
- From: "Maciej W. Rozycki" <macro at imgtec dot com>
- To: <binutils at sourceware dot org>
- Date: Wed, 4 May 2016 01:46:54 +0100
- Subject: [PATCH 2/2] LD/testsuite: Expand STB_GNU_UNIQUE test coverage
- Authentication-results: sourceware.org; auth=none
- References: <alpine dot DEB dot 2 dot 00 dot 1605032355130 dot 6794 at tp dot orcam dot me dot uk>
Expand LD STB_GNU_UNIQUE testing across all eligible targets, as per the
selection made in GAS, that is ones setting the OSABI to ELFOSABI_NONE
or ELFOSABI_GNU.
ld/
* testsuite/ld-unique/unique.exp: Run across all ELFOSABI_NONE
and ELFOSABI_GNU targets
---
Hi,
At first I thought this would be a good idea, prompting architecture
maintainers to fix their code. It wouldn't be over 5 years from commit
f64b2e8d60f2 for me to notice there was something wrong with the MIPS
target if we had any test suite failure for this feature (and then only
because the IFUNC feature currently under review touches related code).
But then I realised there's probably something wrong in the first place
that we require maintainers' intervention for LD whereas we handle this
stuff automagically across all the eligible targets in GAS (see obj-elf.c
and `obj_elf_type' there).
There was an observation made in the original review process here:
<https://sourceware.org/ml/binutils/2011-04/msg00144.html> that we should
be limiting targets by the OSABI, but:
1. GAS does that already in generic code.
2. BFD per-target code doesn't do it anyway.
3. The choice whether to check the OSABI or not is independent from the
choice whether to handle this feature in generic or target-specific BFD
code.
So does anybody actually know why the current per-architecture approach
was accepted in the review?
If not then I'll post a proposal separately to gather all the bits
scattered across target `*_add_symbol_hook' handlers and handle this
feature in `elf_link_add_object_symbols'.
Regardless, OK to apply this test suite change itself?
Maciej
binutils-ld-stb-gnu-unique-test-targets.diff
Index: binutils/ld/testsuite/ld-unique/unique.exp
===================================================================
--- binutils.orig/ld/testsuite/ld-unique/unique.exp 2016-04-30 06:35:48.586138927 +0100
+++ binutils/ld/testsuite/ld-unique/unique.exp 2016-04-30 17:10:41.934056441 +0100
@@ -24,20 +24,20 @@
# Adapted for unique checking by Mark J. Wielaard <mjw@redhat.com>
-# STB_GNU_UNIQUE support has only been implemented for the ix86, x86_64,
-# arm, mips, powerpc, and sparc so far.
-if {!(([istarget "i?86-*-*"]
- || [istarget "x86_64-*-*"]
- || [istarget "arm*-*-*"]
- || [istarget "mips*-*-*"]
- || [istarget "powerpc*-*-*"]
- || [istarget "sparc*-*-*"])
- && ([istarget "*-*-elf*"]
- || [istarget "*-*-nacl*"]
- || (([istarget "*-*-linux*"]
- || [istarget "*-*-gnu*"])
- && ![istarget "*-*-*aout*"]
- && ![istarget "*-*-*oldld*"]))) } {
+# Exclude non-ELF targets.
+if { ![is_elf_format] } {
+ return
+}
+
+# Exclude some more targets; feel free to include your favorite one
+# if you like. The MSP430 and Visium targets set the ELF header's
+# OSABI field to ELFOSABI_STANDALONE and cannot support STB_GNU_UNIQUE.
+if { !([istarget "*-*-elf*"]
+ && ![istarget "msp430-*-*"]
+ && ![istarget "visium-*-*"])
+ && ![istarget *-*-nacl*]
+ && ![istarget *-*-linux*]
+ && ![istarget *-*-gnu*] } {
verbose "UNIQUE tests not run - target does not support UNIQUE"
return
}