Mirror of the gdb-patches mailing list
 help / color / mirror / Atom feed
From: Pedro Alves <pedro@palves.net>
To: gdb-patches@sourceware.org
Subject: [PATCH 03/27] gdb/testsuite: Find host binutils in a native Windows/MSVC config
Date: Thu, 23 Jul 2026 14:00:54 +0100	[thread overview]
Message-ID: <20260723130118.206735-4-pedro@palves.net> (raw)
In-Reply-To: <20260723130118.206735-1-pedro@palves.net>

This is the second of two pieces making plain "make check" Just Work
OOTB when testing a MinGW-hosted GDB targeting the MSVC ABI:

 --host=x86_64-w64-mingw32 --target=x86_64-pc-windows-msvc

The

  gdb_find_objcopy
  gdb_find_objdump
  gdb_find_readelf
  gdb_find_windres
  gdb_find_eu-unstrip

procs in lib/future.exp fall back to dejagnu's "transform", which
prefixes the tool with the target triplet whenever the host and target
triplets differ.

For a MinGW-hosted GDB targeting the MSVC ABI, that fallback produces
e.g. x86_64-pc-windows-msvc-objcopy, which does not exist, so tests
that use these tools fail with, e.g.:

 couldn't execute "x86_64-pc-windows-msvc-objcopy": no such file or directory
 FAIL: gdb.base/step-symless.exp: strip stub symbols

Even though the two triplets differ, this is a Windows native
configuration and a host tool could be used.

Both gdb/configure.ac and gdb/testsuite/configure.ac use
GDB_AC_NATIVE, which treats this --host/--target combination as
native.  gdb/testsuite/configure.ac already stores that result in the
gdb_native substitution variable, so propagate it to the testsuite via
site.exp.  Doing it this way keeps a single source of truth rather
than reimplementing the same logic in Tcl.

Which host tool to run, though, depends on the environment.  For
example:

- The MSYS2 binutils package has bare tools only, e.g. objdump, with
  no x86_64-w64-mingw32-objdump:

  https://packages.msys2.org/packages/mingw-w64-x86_64-binutils

- The xPack MinGW binutils distribution has host-triple-prefixed tools
  only, e.g. x86_64-w64-mingw32-objdump, with no bare objdump:

  https://github.com/xpack-dev-tools/gcc-xpack/

So rather than assume one spelling, search for whichever exists.  Add
a gdb_transform_tool helper that, given a tool name, searches PATH in
the following order:

 1. The target-prefixed name (via "transform").  This is the
    preexisting behavior for a genuine cross.

 2. For native configurations only, the host-triple-prefixed name.

 3. For native configurations only, the bare name.

So for:

 --host=x86_64-w64-mingw32 --target=x86_64-pc-windows-msvc

and for objdump, we try:

  x86_64-pc-windows-msvc-objdump
  x86_64-w64-mingw32-objdump
  objdump

If none of the candidates is found, the helper returns the
target-prefixed name unchanged as before, so the "tool not found"
error is exactly as before.

Change-Id: If48c625503c7c14bf8365e35bfe58c38a3bb7ae9
---
 gdb/testsuite/Makefile.in    |  1 +
 gdb/testsuite/lib/future.exp | 83 +++++++++++++++++++++++++++---------
 gdb/testsuite/lib/gdb.exp    | 13 ++++++
 3 files changed, 77 insertions(+), 20 deletions(-)

diff --git a/gdb/testsuite/Makefile.in b/gdb/testsuite/Makefile.in
index b852dbceb50..c4d7f4f284c 100644
--- a/gdb/testsuite/Makefile.in
+++ b/gdb/testsuite/Makefile.in
@@ -135,6 +135,7 @@ $(abs_builddir)/site.exp site.exp: Makefile
 	echo "set srcdir ${abs_srcdir}" >> ./tmp0; \
 	echo "set tool gdb" >> ./tmp0; \
 	echo "set enable_libctf ${enable_libctf}" >> ./tmp0; \
+	echo "set gdb_native $(if $(filter yes,$(gdb_native)),1,0)" >> ./tmp0; \
 	echo 'source $${srcdir}/lib/append_gdb_boards_dir.exp' >> ./tmp0; \
 	echo "## All variables above are generated by configure. Do Not Edit ##" >> ./tmp0; \
 	cat ./tmp0 > site.exp; \
diff --git a/gdb/testsuite/lib/future.exp b/gdb/testsuite/lib/future.exp
index 3ab160a05fc..ca37fdaaf52 100644
--- a/gdb/testsuite/lib/future.exp
+++ b/gdb/testsuite/lib/future.exp
@@ -146,55 +146,98 @@ proc gdb_find_ldd {} {
     return $ldd
 }
 
+# Return the name of a target binary utility TOOL, e.g. "objcopy".
+#
+# Normally, this is the target-prefixed TOOL, as computed by dejagnu's
+# 'transform'.  On a native configuration, if that name is not on
+# PATH, fall back to a host tool instead.  This makes the binary
+# utilities usable when the host and target triplets differ on what we
+# still consider a native configuration, e.g. a MinGW-hosted GDB
+# targeting the MSVC ABI.
+#
+# Only use this for a tool that behaves the same whether it is the
+# host or the target build of the tool -- the binary utilities that
+# operate on the shared object file format (objcopy, objdump, readelf,
+# etc.) qualify.  Do not use this for a tool whose behavior depends on
+# the target, such as the compiler: the host compiler builds for the
+# host, not for the debuggee's ABI.
+
+proc gdb_transform_tool {tool} {
+    # Compute the target-prefixed name with transform's verbose
+    # logging suppressed -- we may end up only probing this candidate
+    # and not running it (in the native case), so we don't want
+    # transform's "Transforming ..."  line in the log until we know
+    # this is the name we return.
+    set save_verbose $::verbose
+    set ::verbose -1
+    set target_tool [transform $tool]
+    set ::verbose $save_verbose
+
+    # Default to the target-prefixed name.  It is the right tool for a
+    # cross, and for the not-found case it gives the same "tool not
+    # found" error as DejaGnu's transform would.
+    set result $target_tool
+
+    if {[which $target_tool] == 0 && [gdb_is_native_config]} {
+	# Try looking for a host tool: the host-triple-prefixed name
+	# first (e.g. x86_64-w64-mingw32-objcopy), then the bare name
+	# (e.g. objcopy).
+	set host_tool "[ishost {}]-$tool"
+	if {[which $host_tool] != 0} {
+	    set result $host_tool
+	} elseif {[which $tool] != 0} {
+	    set result $tool
+	}
+    }
+
+    # Log the transformation the way dejagnu's transform would, but
+    # for the name we actually settled on.
+    if {$result ne $tool} {
+	verbose "Transforming $tool to $result"
+    }
+
+    return $result
+}
+
 proc gdb_find_objcopy {} {
     global OBJCOPY_FOR_TARGET
     if {[info exists OBJCOPY_FOR_TARGET]} {
-	set objcopy $OBJCOPY_FOR_TARGET
-    } else {
-	set objcopy [transform objcopy]
+	return $OBJCOPY_FOR_TARGET
     }
-    return $objcopy
+    return [gdb_transform_tool objcopy]
 }
 
 # find target objdump
 proc gdb_find_objdump {} {
     global OBJDUMP_FOR_TARGET
     if {[info exists OBJDUMP_FOR_TARGET]} {
-	set objdump $OBJDUMP_FOR_TARGET
-    } else {
-	set objdump [transform objdump]
+	return $OBJDUMP_FOR_TARGET
     }
-    return $objdump
+    return [gdb_transform_tool objdump]
 }
 
 proc gdb_find_readelf {} {
     global READELF_FOR_TARGET
     if {[info exists READELF_FOR_TARGET]} {
-	set readelf $READELF_FOR_TARGET
-    } else {
-	set readelf [transform readelf]
+	return $READELF_FOR_TARGET
     }
-    return $readelf
+    return [gdb_transform_tool readelf]
 }
 
 proc gdb_find_windres {} {
     global WINDRES_FOR_TARGET
     if {[info exists WINDRES_FOR_TARGET]} {
-	set windres $WINDRES_FOR_TARGET
-    } else {
-	set windres [transform windres]
+	return $WINDRES_FOR_TARGET
     }
-    return $windres
+    return [gdb_transform_tool windres]
 }
 
 proc gdb_find_eu-unstrip {} {
     global EU_UNSTRIP_FOR_TARGET
     if {[info exists EU_UNSTRIP_FOR_TARGET]} {
-	set eu_unstrip $EU_UNSTRIP_FOR_TARGET
-    } else {
-	set eu_unstrip [transform eu-unstrip]
+	return $EU_UNSTRIP_FOR_TARGET
     }
-    return $eu_unstrip
+    return [gdb_transform_tool eu-unstrip]
 }
 
 # Local version of default_target_compile, to be used for languages that
diff --git a/gdb/testsuite/lib/gdb.exp b/gdb/testsuite/lib/gdb.exp
index eea39d4e8cc..b4046c402d4 100644
--- a/gdb/testsuite/lib/gdb.exp
+++ b/gdb/testsuite/lib/gdb.exp
@@ -4223,6 +4223,19 @@ proc is_windows_based_target {} {
     return [expr {[istarget *-*-cygwin*] || [istarget *-*-mingw*]}]
 }
 
+# Return true if this is a native configuration in GDB's sense.  This
+# is broader than dejagnu's isnative, which returns true iff we have
+# identical host and target triplets.  On Windows the host and target
+# triplets can differ and the GDB build can still natively debug the
+# target.  E.g. a mingw-hosted GDB targeting binaries produced by a
+# compiler defaulting to windows-msvc.
+
+proc gdb_is_native_config {} {
+    # configure already worked this out and the result is found in
+    # site.exp in this boolean.
+    return $::gdb_native
+}
+
 # Return 1 if displaced stepping is supported on target, otherwise, return 0.
 proc support_displaced_stepping {} {
 
-- 
2.54.0


  parent reply	other threads:[~2026-07-23 13:02 UTC|newest]

Thread overview: 28+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-07-23 13:00 [PATCH 00/27] Teach the testsuite about the Windows/MSVC target Pedro Alves
2026-07-23 13:00 ` [PATCH 01/27] gdb/testsuite: Don't link with -lm on windows-msvc Pedro Alves
2026-07-23 13:00 ` [PATCH 02/27] gdb/testsuite: Use unprefixed runtest for a native Windows/MSVC build Pedro Alves
2026-07-23 13:00 ` Pedro Alves [this message]
2026-07-23 13:00 ` [PATCH 04/27] gdb/testsuite: Compile with -Wno-deprecated-declarations on windows-msvc Pedro Alves
2026-07-23 13:00 ` [PATCH 05/27] gdb/testsuite: Recognize windows-msvc across lib/gdb.exp procedures Pedro Alves
2026-07-23 13:00 ` [PATCH 06/27] gdb/testsuite: Don't pass -fPIC on PE/COFF targets Pedro Alves
2026-07-23 13:00 ` [PATCH 07/27] gdb/testsuite: Use /IMPLIB on windows-msvc to name import libraries Pedro Alves
2026-07-23 13:00 ` [PATCH 08/27] gdb/testsuite: Export all DLL symbols on windows-msvc via generated .def Pedro Alves
2026-07-23 13:01 ` [PATCH 09/27] gdb/testsuite: Restrict --no-as-needed to ELF targets Pedro Alves
2026-07-23 13:01 ` [PATCH 10/27] gdb.base/set-cwd.exp: Use is_windows_native_target Pedro Alves
2026-07-23 13:01 ` [PATCH 11/27] gdb.base/exitsignal.exp: " Pedro Alves
2026-07-23 13:01 ` [PATCH 12/27] gdb.base/exitsignal.exp: Relax SIGSEGV second-chance pattern Pedro Alves
2026-07-23 13:01 ` [PATCH 13/27] gdb/testsuite: Skip -Ttext-segment on PE Pedro Alves
2026-07-23 13:01 ` [PATCH 14/27] gdb/testsuite: Support text_segment on windows-msvc via /BASE Pedro Alves
2026-07-23 13:01 ` [PATCH 15/27] gdb.base/shreloc.exp: Use gdb_compile text_segment to set image base Pedro Alves
2026-07-23 13:01 ` [PATCH 16/27] gdb/coffread: Don't relocate absolute symbols Pedro Alves
2026-07-23 13:01 ` [PATCH 17/27] gdb.base/shreloc.exp: Test absolute symbols portably Pedro Alves
2026-07-23 13:01 ` [PATCH 18/27] gdb: %p => host_address_to_string, target-section owner token Pedro Alves
2026-07-23 13:01 ` [PATCH 19/27] gdb: %p => host_address_to_string, dump_for_expression Pedro Alves
2026-07-23 13:01 ` [PATCH 20/27] gdb: %p => host_address_to_string, find_symtab_matching_filename Pedro Alves
2026-07-23 13:01 ` [PATCH 21/27] gdb: %p => host_address_to_string, handle_output_debug_string Pedro Alves
2026-07-23 13:01 ` [PATCH 22/27] gdb.base/maint-info-sections.exp: Match exec file name with optional .exe Pedro Alves
2026-07-23 13:01 ` [PATCH 23/27] gdb.base/maint-info-sections.exp: Remove stale Windows DATA xfail Pedro Alves
2026-07-23 13:01 ` [PATCH 24/27] gdb.base/solib-weak.exp: Skip on all PE/COFF targets Pedro Alves
2026-07-23 13:01 ` [PATCH 25/27] gdb.server/wrapper.exp: Skip on all Windows targets Pedro Alves
2026-07-23 13:01 ` [PATCH 26/27] gdb/testsuite: Factor out dlopen/LoadLibrary shim into lib/gdb-dlfcn.h Pedro Alves
2026-07-23 13:01 ` [PATCH 27/27] gdb/testsuite/lib/gdb-dlfcn.h: __WIN32__ => _WIN32 Pedro Alves

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20260723130118.206735-4-pedro@palves.net \
    --to=pedro@palves.net \
    --cc=gdb-patches@sourceware.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox