xref: /netbsd-current/doc/HACKS

# $NetBSD: HACKS,v 1.3 2003/04/06 10:12:20 jdc Exp $
#
# This file is intended to document workarounds for currently unsolved
# (mostly) compiler bugs.
#
# Format:
#   hack		title
#   cdate		creation date
#   mdate		mod date
#   who			responsible developer
#   port		...
#  	  affected ports, space separated, if not "all"
#   file		affected file : revision : line from : line to
#  	  affected files and revision and line numbers describing hack
#  	  multiple lines if necessary.
#   pr			NNNN ...
#  	  problem reports this hack works around, if known. Space
#  	  separated.
#   regress		src/regress/directory/where/test/found
#  	  regression test directories, if available.
#   descr
#  	  insert short informal description (multi-line). (Longer ones
#  	  should be in the PR database. More formal descriptions might
#  	  be in the regress tree. See above).
#   kcah
#  	  closing bracket.
#
# this is a comment.

hack	gcc 2.95/vax cannot handle __builtin_ffs()
cdate	Fri Aug 23 21:31:15 CEST 2002
who	ragge
port	vax
file	sys/lib/libkern/libkern.h : 1.42
descr
	__builtin_ffs() is ifdef'd away if __vax__. Matt Thomas has added
	this feature to GCC 3 so it can be removed when switching.
kcah

hack	gcc 2.95.3 -O2 bug
cdate	Sun Jan  6 23:17:28 UTC 2002
who	lukem
port	alpha
file	sbin/newfs/mkfs.c : (all?)
pr	15156
descr
	newfs built with gcc -O2 dumps core part-way through.
	using -O1 avoids the problem
kcah

hack	binutil-2.11-sparc64-pltrela
mdate	14 Aug 2001
who	eeh
port	sparc64
file	libexec/ld.elf_so/reloc.c: 1.41 : 597 : 619
descr
	The first for PLT entries are reserved.  There is some disagreement
	whether they should have associated relocation entries.  Both the
	SPARC 32-bit and 64-bit ELF specifications say that they should have
	relocation entries, but the 32-bit SPARC binutils do not generate
	them, and now the 64-bit SPARC binutils have stopped generating them
	too.

	To provide binary compatibility, we will check the first entry, if it
	is reserved it should not be of the type JMP_SLOT.  If it is JMP_SLOT,
	then the 4 reserved entries were not generated and our index is 4
	entries too far, so we frob the rela pointer.
kcah

hack	gcc-arm32-schedule
mdate	04 May 2000
who	is
file	lib/libc/stdlib/strtoull.c	: 1.3 : 78 : 81
file	lib/libc/stdlib/strtouq.c	: 1.15 : 73 : 76
pr	9613
descr
	strtoull() was returning a wrong result for small numbers with bit 31
	set. This is a gcc/arm32 compiler bug in gcc < 2.95.2. Adding
	(void)&acc; works around it.
kcah

hack	gcc-sparc64-memorysize
mdate	17 Sep 2000
who	mrg
port	sparc64
file	lib/libc/hash/sha1.c		: 1.6 : 78 : 129 : 162 : 189
file	sys/lib/libkern/sha1.c		: 1.6 : 78 : 129 : 162 : 189
file	usr.sbin/bind/lib/Makefile	: 1.14 : 87 : 91
descr
	the normal SHA1Transform() function causes the sparc64 compiler to
	use excessive amounts of memory attempting to optimise this function.
	using no optimisation allows the function to build.  this hack
	splits the mathematical functions inside SHA1Transform() into 4
	separate functions, which avoids the optimiser explosion.
kcah

hack	gcc-sparc64-double-arguments
mdate	18 Sep 2000
who	mrg
port	sparc64
file	xsrc/xc/lib/fonts/Type1/type1.c		: 1.2 : 1461 : 1507 : 1702 : 1709
descr
	in passing 17 "double" arguments to a function, gcc incorrectly tries
	to use a register that doesn't exist, causing an internal error.  this
	hack changes the way arguments are passed to the function, such that
	a pointer to a "double" (really an array) is passed, and the old
	arguments are assigned within the function.
kcah

hack	gcc-sparc64-gawk
mdate	28 Oct 2000
who	mrg
port	sparc64
file	gnu/usr.bin/gawk/Makefile		: 1.11 : 21 : 24
descr
	gcc fails to build gawk properly, due to double floating point bugs.
	this simple test program will fail to print ``ab'':
		BEGIN { f = "abc"; print substr(f, 1, 2); exit }
	the work around is simple: use -msoft-quad-float.
kcah

hack    gcc-sparc64-libgcc
mdate   01 Dec 2001
who	jmc
port	sparc64
file	gnu/lib/libgcc/Makefile			: 1.29 : 24 : 27
descr
	gcc fails to do double -> int conversions correctly without 
	soft-quad-float. Add it in there for the internal conversion
	routines. This appears to be more than likely emulations bugs than
	compiler bugs as the assembly looks correct on the surface.

kcah

hack	gcc-sparc64-iostat-systat
mdate	04 Feb 2001
who	mrg
port	sparc64
file	usr.sbin/iostat/Makefile	: 1.17 : 7 : 10
file	usr.bin/systat/Makefile		: 1.23 : 6 : 9
descr
	gcc fails to build iostat properly, due to double floating point bugs.
	results will include "Inf" and other garbage.
	the work around is simple: use -msoft-quad-float.
kcah

hack	gcc-sparc64-nawk
mdate	25 Oct 2001
who	martin
port	sparc64
file	usr.bin/awk/Makefile		: 1.3 : 13 : 15
descr
	Don't use long double on sparc64.
kcah

hack	gcc-sparc64-ntp
mdate	12 Nov 2000
who	martin
port	sparc64
file	usr.sbin/ntp/Makefile.inc	: 1.5 : 17 : 19
descr
	gcc fails to build ntpd properly, due to double floating point bugs.
	Ntpd will fail to decode incoming packets; ntpdc -s will report
	a delay, offset and dispersion of 0.0000 for all peers.
	the work around is simple: use -msoft-quad-float.
kcah

hack	gcc-sparc64-x_dhclient
mdate	27 Dec 2000
who	mrg
port	sparc64
file	distrib/utils/x_dhclient/Makefile	: 1.6 : 28 : 30
descr
	gcc fails with an ICE while building dhclient's parse.c with -O2
	and -Os.  the work around is simple: don't use -Os.
kcah

hack	lint-no-long-double
mdate	17 Aug 2001
who	eeh
port	sparc64
file	usr.bin/xlint/lint1/param.h: 1.14 :
descr
	Don't use long double on sparc64.
kcah

hack	netstat ieee1394 address printing.
mdate	14 Nov 2000
who	matt
file	lib/libc/net/getnameinfo.c	: 1.32 : 497 : 503
descr
	Because the current implementation of IP over IEEE1394, the
	fw device address contains more than just the IEEE1394 EUI-64.
	So when printing out IEEE1394 addresses, ignore the extra stuff.
kcah

hack	xterm vs. libterm
mdate	01 Aug 2000
who	jdc
file	xsrc/xc/programs/xterm/main.c	: 1.2 : 3609 : 3614
pr	10383
descr
	In order to extend the termcap string over 1023 bytes, a ZZ entry was
	introduced to point to a memory location containing the full entry.
	Without this hack, xterm will export a termcap containing the ZZ
	entry, which will then be ignored by libterm.  As xterm modifies the
	exported termcap, this would cause those modifications to be ignored.
kcah

hack	gcc-3-libtelnet
mdate	29 Jan 2002
who	thorpej
file	lib/libtelnet/Makefile	: 1.14 : 14 : 17
descr
	GCC 3.x issues a spurious warning when compiling a Duff's Device
	with optimization, even though it compiles the file correctly.
	GCC PR optimization/5230.  kerberos.c contains a Duff's Device
	in the checksum routine.  Hack is to build libtelnet w/o optimization
	if HAVE_GCC3 is set.
kcah

hack	gcc-sh3-sed
mdate	23 Apr 2002
who	thorpej
file	usr.bin/sed/Makefile	: 1.9 : 9 : 13
descr
	The in-tree GCC 2.95.3-based compiler ICEs when building
	with optimization for SuperH.  Hack is to biuld with -O0.
kcah

hack	gcc-unsigned-compare
cdate	09 Mar 2002
mdate	18 Mar 2002
who	bjh21
port	arm
file	dist/bind/lib/nameser/ns_parse.c : 1.3
file	dist/dhcp/minires/ns_parse.c : 1.3
file	dist/dhcp/omapip/result.c : 1.2
file	dist/dhcp/server/failover.c : 1.3
file	gnu/dist/gawk/eval.c : 1.4
file	gnu/dist/toolchain/bfd/bfd.c : 1.2
file	gnu/dist/toolchain/bfd/format.c : 1.2
file	gnu/dist/toolchain/gdb/target.c : 1.2
file	sys/kern/vfs_subr.c : 1.172
descr	When checking that a potentially-unsigned enum is >= 0, assign it
	to an int first.  This is necessary to avoid "comparison is always
	true" warnings with -fshort-enums.  Casting to an int really should
	be enough, but turns out not to be.
kcah

hack	wi-at-big-endian-bus
cdate	15 Mar 2002
who	martin
file	dev/ic/wireg.h
descr	Add an option to access the underlying bus in big endian byte order
	to work around deficiencies in bus_space_{read,write}_* macros.
	Those don't allow the implementation of a proper pcmcia bus space
	tag.
kcah

hack	gcc 2.95.3 -O2 (-fgcse) bug
cdate	Sun May  5 18:36:04 UTC 2002
who	tsutsui
port	macppc
file	sys/arch/macppc/dev/adb_direct.c:	1.24 : 1895 : 1896
pr	16678
descr
	gcc-2.95.3 does generates wrong code on optimization
	by gcc -O2 (-fgcse), and adb_read_date_time() returns
	wrong value on cuda system.
	A null asm statement has been added to avoid this for workaround.
kcah

hack	gcc-vax-libbz2
mdate	27 Jun 2002
who	thorpej
port	vax
file	lib/libbz2/Makefile
descr
	libbz2 is mis-compiled with optimization with GCC 2.95.3
	on VAX.  -O0 works around this problem.
kcah

hack	egcs-pc532-ip6_mroute
cdate	09 Jul 2002
who	simonb
port	pc532
file	sys/arch/pc532/conf/Makefile.pc532 : 1.70
file	sys/arch/pc532/conf/files.pc532 : 1.47
descr
	egcs 1.1.2 gets an "internal error--insn does not satisfy its
	constraints" error compiling ip6_mroute.c with -O2 or greater.
	-O1 works around this problem.
kcah

hack	SuperH SH5 Toolchain Bugs
cdate	11 Jul 2002
who	scw
port	sh5
file	usr.sbin/ndbootd/ndbootd.c : 1.6
file	usr.sbin/traceroute/traceroute.c : 1.48
descr
	The SuperH SH5 toolchain (2.97-sh5-010522) gets an internal
	compiler error when assigning a bit-wise inverted value
	under some circumstances.
	Work around it by splitting the statement into two.
kcah

hack	vax Toolchain bug
cdate	28 Mar 2003
who	he
port	vax
file	sys/arch/vax/vax/intvec.S : 1.5
descr
	Workaround for PR toolchain/20924.  The assembler apparently
	tries to range-check byte offsets when it doesn't have
	sufficient information to make that decision.
	Workaround uses `brw' instead of `brb' instruction.
kcah

hack	specific knowledge of colours in curses code
cdate	Sun Apr  6 11:05:24 BST 2003
who	jdc
port	all
file	lib/libcurses/color.c : r1.24
descr
	Swap red/blue and yellow/cyan colours for COLOR_OTHER.
	Fix is to enhance libtermcap to understand terminfo-style % sequences.
	See also:
	    http://mail-index.netbsd.org/tech-userlevel/2003/04/06/0000.html
kcah