From: Simon Marchi <simon.marchi@polymtl.ca>
Date: Mon, 13 Jan 2020 18:59:18 +0000 (-0500)
Subject: gdb: make regformats output a declaration for the init function
X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=e2de1eec22358c64510dc18e755d0ff4cf91ad7a;p=binutils-gdb.git

gdb: make regformats output a declaration for the init function

When compiling gdbserver for an architecture that uses the regdat.sh
script (such as m68k) and the -Wmissing-declarations compiler flag, I
get:

  REGDAT reg-m68k-generated.c
  CXX    reg-m68k.o
reg-m68k-generated.c:30:1: error: no previous declaration for 'void init_registers_m68k()' [-Werror=missing-declarations]
   30 | init_registers_m68k (void)
      | ^~~~~~~~~~~~~~~~~~~

The same happens with other architectures, such as s390, but I'll be
using 68k as an example.

The init_registers_m68k function is defined in reg-m68k-generated.c,
which is produced by the regformats/regdat.sh script.  This script reads
the regformats/reg-m68k.dat file, containing a register description, and
produces C code that creates a corresponding target description at
runtime.

The init_registers_m68k function is invoked at initialization time in
linux-m68k-low.c.  The function must therefore be non-static, but does
not have a declaration at the moment.

The real clean way of fixing this would be to make regdat.sh generate a
.h file (in addition to the .c file) with declarations for whatever is
in the .c file.  The generated .c file would include the .h file, and
therefore the definition would have a corresponding declaration.  The
linux-m68k-low.c file would also include this .h file, instead of having
its own declaration of init_registers_m68k, like it does now.

However, this would be a quite big change for not much gain.  As far as
I understand, some common architectures (i386, x86-64, ARM, AArch64)
have been moved to dynamically building target descriptions based on
features (the linux-*-tdesc.c files in gdbserver) and don't use
regdat.sh anymore.  Logically (and given infinite development
resources), the other architectures would be migrated to this system too
and the regdat.sh script would be dropped.  A new architecture would
probably not use regdat.sh either.  So I therefore propose this simpler
patch instead, which just adds a local declaration in the generated
file.

gdb/ChangeLog:

	* regformats/regdat.sh: Generate declaration for init function.
---

diff --git a/gdb/ChangeLog b/gdb/ChangeLog
index 64b64fb242d..78387ff010a 100644
--- a/gdb/ChangeLog
+++ b/gdb/ChangeLog
@@ -1,3 +1,7 @@
+2020-01-13  Simon Marchi  <simon.marchi@polymtl.ca>
+
+	* regformats/regdat.sh: Generate declaration for init function.
+
 2020-01-13  Simon Marchi  <simon.marchi@polymtl.ca>
 
 	* remote-sim.c (next_pid, INITIAL_PID, sim_inferior_data): Move
diff --git a/gdb/regformats/regdat.sh b/gdb/regformats/regdat.sh
index 1839a881213..a40f2336484 100755
--- a/gdb/regformats/regdat.sh
+++ b/gdb/regformats/regdat.sh
@@ -127,6 +127,10 @@ do
 
     echo "const struct target_desc *tdesc_${name};"
     echo ""
+
+    # This is necessary for -Wmissing-declarations.
+    echo "void init_registers_${name} (void);"
+
     echo "void"
     echo "init_registers_${name} (void)"
     echo "{"