[PATCH] PR gas/20312: Do not pad sections to alignment on failed assembly

["Maciej W. Rozycki" <macro at imgtec dot com>]

Correct a regression from commit 85024cd8bcb9 ("Run write_object_file 
after errors") causing unsuccessful assembly, which may be due to any 
reason, such as supplying a valid source like this:

	.text
	.byte	0
	.err

to terminate with an assertion failure like:

test.s: Assembler messages:
test.s:3: Error: .err encountered
../as-new: BFD (GNU Binutils) 2.24.51.20140628 internal error, aborting at .../gas/write.c line 608 in size_seg
../as-new: Please report this bug.

on targets whose default text section alignment is above 0, typically 
RISC machines.

This is due to an attempt to set last text section's frag alignment to 
0, requested from `subsegs_finish_section' where `frag_align_code 
(alignment, 0)' is called with `alignment' set to 0 rather than the 
section alignment if `had_errors' has returned true.  The call to 
`subsegs_finish_section' is made from `subsegs_finish' from 
`write_object_file' at unsuccessful completion, which previously wasn't 
made.

Always set last section's frag alignment from the section alignment 
then, forcing no section padding instead if completing unsuccessfully, 
so that in that case alignment padding is still suppressed from any 
listing generated, fixing assertion failures for these targets:

alpha-linuxecoff  -FAIL: all pr20312
arm-aout  -FAIL: all pr20312
mips-freebsd  -FAIL: all pr20312
mips-img-linux  -FAIL: all pr20312
mips-linux  -FAIL: all pr20312
mips-mti-linux  -FAIL: all pr20312
mips-netbsd  -FAIL: all pr20312
mips-sgi-irix5  -FAIL: all pr20312
mips-sgi-irix6  -FAIL: all pr20312
mips-vxworks  -FAIL: all pr20312
mips64-freebsd  -FAIL: all pr20312
mips64-img-linux  -FAIL: all pr20312
mips64-linux  -FAIL: all pr20312
mips64-mti-linux  -FAIL: all pr20312
mips64-openbsd  -FAIL: all pr20312
mips64el-freebsd  -FAIL: all pr20312
mips64el-img-linux  -FAIL: all pr20312
mips64el-linux  -FAIL: all pr20312
mips64el-mti-linux  -FAIL: all pr20312
mips64el-openbsd  -FAIL: all pr20312
mipsel-freebsd  -FAIL: all pr20312
mipsel-img-linux  -FAIL: all pr20312
mipsel-linux  -FAIL: all pr20312
mipsel-mti-linux  -FAIL: all pr20312
mipsel-netbsd  -FAIL: all pr20312
mipsel-vxworks  -FAIL: all pr20312
mipsisa32-linux  -FAIL: all pr20312
mipsisa32el-linux  -FAIL: all pr20312
mipsisa64-linux  -FAIL: all pr20312
mipsisa64el-linux  -FAIL: all pr20312
sh-pe  -FAIL: all pr20312
sparc-aout  -FAIL: all pr20312

	gas/
	PR gas/20312
	* write.c (subsegs_finish_section): Force no section padding to
	alignment on failed assembly, always set last frag's alignment 
	from section.
	* testsuite/gas/all/pr20312.l: New list test.
	* testsuite/gas/all/pr20312.s: New test source.
	* testsuite/gas/all/gas.exp: Run the new test
---
 [Nick, thanks for adding `--no-pad-sections' at the right moment, it made 
this fix very easy.]

 Regression tested across the usual set of 162 targets.  OK to apply?

  Maciej

binutils-gas-subseg-align-error.diff
Index: binutils/gas/testsuite/gas/all/gas.exp
===================================================================
--- binutils.orig/gas/testsuite/gas/all/gas.exp	2016-06-29 01:00:54.050624527 +0100
+++ binutils/gas/testsuite/gas/all/gas.exp	2016-06-29 01:00:56.940675554 +0100
@@ -448,6 +448,8 @@ if [is_elf_format] {
 
 run_dump_test quoted-sym-names
 
+run_list_test pr20312
+
 load_lib gas-dg.exp
 dg-init
 dg-runtest [lsort [glob -nocomplain $srcdir/$subdir/err-*.s $srcdir/$subdir/warn-*.s]] "" ""
Index: binutils/gas/testsuite/gas/all/pr20312.l
===================================================================
--- /dev/null	1970-01-01 00:00:00.000000000 +0000
+++ binutils/gas/testsuite/gas/all/pr20312.l	2016-06-29 01:00:56.957784759 +0100
@@ -0,0 +1,2 @@
+.*: Assembler messages:
+.*:3: Error: \.err encountered
Index: binutils/gas/testsuite/gas/all/pr20312.s
===================================================================
--- /dev/null	1970-01-01 00:00:00.000000000 +0000
+++ binutils/gas/testsuite/gas/all/pr20312.s	2016-06-29 01:00:56.970850699 +0100
@@ -0,0 +1,3 @@
+	.text
+	.byte	0
+	.err
Index: binutils/gas/write.c
===================================================================
--- binutils.orig/gas/write.c	2016-06-29 00:55:28.632856532 +0100
+++ binutils/gas/write.c	2016-06-29 01:00:25.608473607 +0100
@@ -1739,24 +1739,24 @@ subsegs_finish_section (asection *s)
       /* This now gets called even if we had errors.  In that case,
 	 any alignment is meaningless, and, moreover, will look weird
 	 if we are generating a listing.  */
-      if (!had_errors ())
-	{
-	  alignment = SUB_SEGMENT_ALIGN (now_seg, frchainP);
-	  if ((bfd_get_section_flags (now_seg->owner, now_seg) & SEC_MERGE)
-	      && now_seg->entsize)
-	    {
-	      unsigned int entsize = now_seg->entsize;
-	      int entalign = 0;
+      if (had_errors ())
+	do_not_pad_sections_to_alignment = 1;
 
-	      while ((entsize & 1) == 0)
-		{
-		  ++entalign;
-		  entsize >>= 1;
-		}
+      alignment = SUB_SEGMENT_ALIGN (now_seg, frchainP);
+      if ((bfd_get_section_flags (now_seg->owner, now_seg) & SEC_MERGE)
+	  && now_seg->entsize)
+	{
+	  unsigned int entsize = now_seg->entsize;
+	  int entalign = 0;
 
-	      if (entalign > alignment)
-		alignment = entalign;
+	  while ((entsize & 1) == 0)
+	    {
+	      ++entalign;
+	      entsize >>= 1;
 	    }
+
+	  if (entalign > alignment)
+	    alignment = entalign;
 	}
 
       if (subseg_text_p (now_seg))