On 16 Apr 2007 13:00:27 -0000
Parrot Bug Summary <[EMAIL PROTECTED]> wrote:
> 41858 [CAGE] Make a reasonable set of rules for splint
Hi,
Random cage cleaner here. At Andy Lester's suggestion, I've begun
working on this splint stuff. So everything below is all his fault. :)
I found that "make splint" is currently very dependent on the machine
you run it on. I'm running it on Gentoo Linux, and so, Parrot gets
built with lots of unixish features (due to their being enabled in
has_header.h). For instance, my platform has PARROT_HAS_HEADER_ARPAINET
defined, so <arpa/inet.h> gets included.
Unfortunately, splint barfs on this, because my arpa/inet.h isn't
compatible with the +posixstrictlibs flag (because of a lack of
in_addr_t typedef in splint's posix lib, I think). Further on, I found
splint couldn't parse my pthread.h either, so I eventually gave it a
+skip-sys-headers, which got a lot farther.
I battled against this for a while, and found this small set of changes
(see the attached diff) were helpful to get the majority of the
sourcefiles parsing correctly with splint.
Unfortunately, I had to turn +posixstrictlibs into +unixstrictlibs to
get it to work at all on my machine... which is not the most portable
thing to do.
Do you think its possible to use a "dummy" config.h and has_header.h
and stuff, for the purposes of splint portability? One that doesn't
enable any platform-specific stuff, and thus, will generate the same
results on any platform? Splint seems to like using its own internal
header-library (and it has a few different ones... ISO, POSIX, UNIX),
and barfs on my system headers in several places, mostly due to
conflicts between its library and my reality.
I had to blacklist the following files, because splint could not parse
them for whatever reason:
compilers/imcc/imcparser.c
compilers/imcc/instructions.c
compilers/imcc/parser_util.c
src/charset/unicode.c
src/encodings/ucs2.c
src/encodings/utf16.c
src/exec_save.c
src/io/io.c
src/misc.c
src/ops/core_ops_cg.c
src/ops/core_ops_cgp.c
src/pmc/bigint.c
src/spf_render.c
src/spf_vtable.c
src/string_primitives.c
But the remaining 177 C source files went through okay, with the
following warning summary:
Error Type Reported Suppressed
=================== ======== =========
nullderef 0 512
nullpass 0 350
nullret 0 617
nullstate 0 363
nullassign 0 148
usedef 0 170
compdef 0 3858
initallelements 13 0
incondefs 1 0
realcompare 224 21
unsignedcompare 0 17
nullptrarith 0 7
bitwisesigned 4354 58
shiftnegative 0 78
shiftimplementation 0 764
sizeoftype 0 2181
fixedformalarray 4 0
formalarray 0 22
boolops 0 1286
predboolint 0 14
predboolptr 0 1096
predboolothers 837 0
abstract 0 9
mustfreefresh 0 3229
mustfreeonly 0 4737
usereleased 0 138
compdestroy 0 24010
branchstate 0 958
compmempass 0 1681
stackref 9 2
dependenttrans 0 23
onlytrans 0 719
onlyunqglobaltrans 0 3
temptrans 0 3409
kepttrans 0 248
immediatetrans 0 49
statictrans 0 41
unqualifiedtrans 0 419
staticinittrans 0 2
readonlytrans 0 721
aliasunique 1 0
mayaliasunique 0 8
observertrans 0 35
modobserver 2 0
globs 0 271
evalorder 5 0
evalorderuncon 0 131
casebreak 0 1251
misscase 0 1
looploopbreak 0 35
loopswitchbreak 0 430
switchswitchbreak 0 44
looploopcontinue 0 14
ifempty 3 0
noeffect 2 0
retvalother 478 103
retvalint 0 513
macroredef 88 0
macroparens 0 1
sefuncon 0 5
namechecks 157 167
isoreserved 0 1008
fcnuse 78 0
paramuse 0 17175
varuse 38 3
declundef 196 0
exitarg 0 9
shadow 22 85
nestedextern 22 0
redecl 0 191
redef 8 5
exporttype 0 226
type 0 8620
formatconst 0 1
voidabstract 0 1
castfcnptr 0 168
charindex 0 10
ignoresigns 2 0
unrecog 61 0
bufferoverflowhigh 28 0
multithreaded 0 1
superuser 0 1
warnlintcomments 0 1
======== =========
Total 6633 82474
I think this is a reasonable starting point. More details are available
on request. (the stdout was around 2 megs.)
So far, I've already identified the "casebreak" warnings as useless,
because parrot often intentionally falls through from one switchcase to
another, and this isn't a bug. So I shut those off. It's likely that
other flags in the above list will need to be suppressed, as well...
the casebreak one was simply the most obvious one.
I'd like to spend some time trying to make "make splint" more
portable... splint defines a special preprocessor variable when it
runs, so maybe config.h can skip all the platform-specific stuff when
that variable is defined, or something like that. And I'd like to
spend some time fixing the source files I listed above, or, failing
that, marking them somehow so splint will skip the erroneous stuff.
I'd also like to spend some time researching and (hopefully) fixing
some of the warnings in the above warning list, too.
Is anyone already working on this?
Does anyone have any interest in this?
Does anyone have an objection to it?
Mark
=== config/gen/makefiles/root.in
==================================================================
--- config/gen/makefiles/root.in (revision 20093)
+++ config/gen/makefiles/root.in (local)
@@ -126,15 +126,15 @@
SPLINT = splint
# note: +distinctinternalnames with +externalnamelen at 8 characters
# (the ANSI standard) is not going to fly for us.
-SPLINTFLAGS = -weak +showscan +showsummary +posixstrictlib -abstract \
+SPLINTFLAGS = -weak +showscan +showsummary +unixstrictlib -abstract \
+aliasunique +ansireserved +ansireservedinternal +assignexpose \
+bitwisesigned +boolcompare +booltype BOOLVAL +casebreak -castfcnptr \
-charint +continuecomment +controlnestdepth 15 +cppnames +declundef \
+distinctinternalnames +evalorder +fielduse +forcehints -globs \
- +imptype +includenest 8 +incondefs +incondefslib +linelen 9999 \
+ +imptype +includenest 20 +incondefs +incondefslib +linelen 9999 \
+longintegral +macroassign +macroempty +macroredef +matchanyintegral \
+nestedextern +noeffect +readonlystrings +realcompare +shadow -type \
- -isoreserved
+ -isoreserved +skip-sys-headers +single-include -casebreak
# slightly weaker, more realistic variant.
SPLINTFLAGS2 = $(SPLINTFLAGS) -ansireserved -retvalother -bitwisesigned \
