diff options
author | Larry Wall <lwall@netlabs.com> | 1991-11-05 06:26:52 +0000 |
---|---|---|
committer | Larry Wall <lwall@netlabs.com> | 1991-11-05 06:26:52 +0000 |
commit | 11aea3600896e20487883b2cb767b57027617482 (patch) | |
tree | 6c185e72ae0018f8cda24d7c5584596eda70c555 /c2ph.doc | |
parent | f1ca563b1a195a92d64b84ed3b4f859a5807e204 (diff) | |
download | perl-11aea3600896e20487883b2cb767b57027617482.tar.gz |
perl 4.0 patch 12: patch #11, continued
See patch #11.
Diffstat (limited to 'c2ph.doc')
-rw-r--r-- | c2ph.doc | 191 |
1 files changed, 191 insertions, 0 deletions
diff --git a/c2ph.doc b/c2ph.doc new file mode 100644 index 0000000000..0c3eaee7eb --- /dev/null +++ b/c2ph.doc @@ -0,0 +1,191 @@ +Article 484 of comp.lang.perl: +Xref: netlabs comp.lang.perl:484 comp.lang.c:983 alt.sources:134 +Path: netlabs!psinntp!iggy.GW.Vitalink.COM!lll-winken!sun-barr!cronkite.Central.Sun.COM!spdev!texsun!convex!tchrist +From: tchrist@convex.com (Tom Christiansen) +Newsgroups: comp.lang.perl,comp.lang.c,alt.sources +Subject: pstruct -- a C structure formatter; AKA c2ph, a C to perl header translator +Keywords: C perl tranlator +Message-ID: <1991Jul25.081021.8104@convex.com> +Date: 25 Jul 91 08:10:21 GMT +Sender: usenet@convex.com (news access account) +Followup-To: comp.lang.perl +Organization: CONVEX Computer Corporation, Richardson, Tx., USA +Lines: 1208 +Nntp-Posting-Host: pixel.convex.com + +Once upon a time, I wrote a program called pstruct. It was a perl +program that tried to parse out C structures and display their member +offsets for you. This was especially useful for people looking at +binary dumps or poking around the kernel. + +Pstruct was not a pretty program. Neither was it particularly robust. +The problem, you see, was that the C compiler was much better at parsing +C than I could ever hope to be. + +So I got smart: I decided to be lazy and let the C compiler parse the C, +which would spit out debugger stabs for me to read. These were much +easier to parse. It's still not a pretty program, but at least it's more +robust. + +Pstruct takes any .c or .h files, or preferably .s ones, since that's +the format it is going to massage them into anyway, and spits out +listings like this: + +struct tty { + int tty.t_locker 000 4 + int tty.t_mutex_index 004 4 + struct tty * tty.t_tp_virt 008 4 + struct clist tty.t_rawq 00c 20 + int tty.t_rawq.c_cc 00c 4 + int tty.t_rawq.c_cmax 010 4 + int tty.t_rawq.c_cfx 014 4 + int tty.t_rawq.c_clx 018 4 + struct tty * tty.t_rawq.c_tp_cpu 01c 4 + struct tty * tty.t_rawq.c_tp_iop 020 4 + unsigned char * tty.t_rawq.c_buf_cpu 024 4 + unsigned char * tty.t_rawq.c_buf_iop 028 4 + struct clist tty.t_canq 02c 20 + int tty.t_canq.c_cc 02c 4 + int tty.t_canq.c_cmax 030 4 + int tty.t_canq.c_cfx 034 4 + int tty.t_canq.c_clx 038 4 + struct tty * tty.t_canq.c_tp_cpu 03c 4 + struct tty * tty.t_canq.c_tp_iop 040 4 + unsigned char * tty.t_canq.c_buf_cpu 044 4 + unsigned char * tty.t_canq.c_buf_iop 048 4 + struct clist tty.t_outq 04c 20 + int tty.t_outq.c_cc 04c 4 + int tty.t_outq.c_cmax 050 4 + int tty.t_outq.c_cfx 054 4 + int tty.t_outq.c_clx 058 4 + struct tty * tty.t_outq.c_tp_cpu 05c 4 + struct tty * tty.t_outq.c_tp_iop 060 4 + unsigned char * tty.t_outq.c_buf_cpu 064 4 + unsigned char * tty.t_outq.c_buf_iop 068 4 + (*int)() tty.t_oproc_cpu 06c 4 + (*int)() tty.t_oproc_iop 070 4 + (*int)() tty.t_stopproc_cpu 074 4 + (*int)() tty.t_stopproc_iop 078 4 + struct thread * tty.t_rsel 07c 4 + + etc. + + +Actually, this was generated by a particular set of options. You can control +the formatting of each column, whether you prefer wide or fat, hex or decimal, +leading zeroes or whatever. + +All you need to be able to use this is a C compiler than generates +BSD/GCC-style stabs. The -g option on native BSD compilers and GCC +should get this for you. + +To learn more, just type a bogus option, like -\?, and a long usage message +will be provided. There are a fair number of possibilities. + +If you're only a C programmer, than this is the end of the message for you. +You can quit right now, and if you care to, save off the source and run it +when you feel like it. Or not. + + + +But if you're a perl programmer, then for you I have something much more +wondrous than just a structure offset printer. + +You see, if you call pstruct by its other incybernation, c2ph, you have a code +generator that translates C code into perl code! Well, structure and union +declarations at least, but that's quite a bit. + +Prior to this point, anyone programming in perl who wanted to interact +with C programs, like the kernel, was forced to guess the layouts of the C +strutures, and then hardwire these into his program. Of course, when you +took your wonderfully to a system where the sgtty structure was laid out +differently, you program broke. Which is a shame. + +We've had Larry's h2ph translator, which helped, but that only works on +cpp symbols, not real C, which was also very much needed. What I offer +you is a symbolic way of getting at all the C structures. I've couched +them in terms of packages and functions. Consider the following program: + + #!/usr/local/bin/perl + + require 'syscall.ph'; + require 'sys/time.ph'; + require 'sys/resource.ph'; + + $ru = "\0" x &rusage'sizeof(); + + syscall(&SYS_getrusage, &RUSAGE_SELF, $ru) && die "getrusage: $!"; + + @ru = unpack($t = &rusage'typedef(), $ru); + + $utime = $ru[ &rusage'ru_utime + &timeval'tv_sec ] + + ($ru[ &rusage'ru_utime + &timeval'tv_usec ]) / 1e6; + + $stime = $ru[ &rusage'ru_stime + &timeval'tv_sec ] + + ($ru[ &rusage'ru_stime + &timeval'tv_usec ]) / 1e6; + + printf "you have used %8.3fs+%8.3fu seconds.\n", $utime, $stime; + + +As you see, the name of the package is the name of the structure. Regular +fields are just their own names. Plus the follwoing accessor functions are +provided for your convenience: + + struct This takes no arguments, and is merely the number of first-level + elements in the structure. You would use this for indexing + into arrays of structures, perhaps like this + + + $usec = $u[ &user'u_utimer + + (&ITIMER_VIRTUAL * &itimerval'struct) + + &itimerval'it_value + + &timeval'tv_usec + ]; + + sizeof Returns the bytes in the structure, or the member if + you pass it an argument, such as + + &rusage'sizeof(&rusage'ru_utime) + + typedef This is the perl format definition for passing to pack and + unpack. If you ask for the typedef of a nothing, you get + the whole structure, otherwise you get that of the member + you ask for. Padding is taken care of, as is the magic to + guarantee that a union is unpacked into all its aliases. + Bitfields are not quite yet supported however. + + offsetof This function is the byte offset into the array of that + member. You may wish to use this for indexing directly + into the packed structure with vec() if you're too lazy + to unpack it. + + typeof Not to be confused with the typedef accessor function, this + one returns the C type of that field. This would allow + you to print out a nice structured pretty print of some + structure without knoning anything about it beforehand. + No args to this one is a noop. Someday I'll post such + a thing to dump out your u structure for you. + + +The way I see this being used is like basically this: + + % h2ph <some_include_file.h > /usr/lib/perl/tmp.ph + % c2ph some_include_file.h >> /usr/lib/perl/tmp.ph + % install + +It's a little tricker with c2ph because you have to get the includes right. +I can't know this for your system, but it's not usually too terribly difficult. + +The code isn't pretty as I mentioned -- I never thought it would be a 1000- +line program when I started, or I might not have begun. :-) But I would have +been less cavalier in how the parts of the program communicated with each +other, etc. It might also have helped if I didn't have to divine the makeup +of the stabs on the fly, and then account for micro differences between my +compiler and gcc. + +Anyway, here it is. Should run on perl v4 or greater. Maybe less. + + +--tom + + |