Using lex in a shared library

[Matthew Fleming]

On Fri, Jul 2, 2010 at 11:54 PM, Xin LI <delphij at delphij.net> wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA256
>
> On 2010/07/02 16:52, Xin LI wrote:
>> On 2010/07/02 16:34, Matthew Fleming wrote:
>>> On Fri, Jul 2, 2010 at 4:02 PM, Garrett Cooper <yanefbsd at gmail.com> wrote:
>>>> On Fri, Jul 2, 2010 at 2:51 PM, Matthew Fleming <mdf356 at gmail.com> wrote:
>>>>> I have the following Makefile for a shared library at $work:
>>>>>
>>>>> ISI_TOP=        ../..
>>>>>
>>>>> LIB=            isi_date
>>>>> SHLIB_MAJOR=    1
>>>>> SHLIB_MINOR=    0
>>>>> SRCS=           date.c date_parser.new.c lex.yy.c
>>>>> INCS=           date.h
>>>>> INCLUDEDIR=     /usr/include/isi_date
>>>>>
>>>>> YFLAGS+=        -vt
>>>>> FLEX=           /usr/bin/flex
>>>>> LDADD=          -ll
>>>>>
>>>>> CLEANFILES+=    date_parser.new.c y.tab.h y.tab.c lex.yy.c y.output \
>>>>>                check_date.log test
>>>>>
>>>>> lex.yy.c: date_lexer.new.l
>>>>>        ${FLEX} $>
>>>>>
>>>>> CFLAGS+=        -I${.CURDIR}
>>>>> #CFLAGS+=       -g
>>>>>
>>>>> .include "${ISI_TOP}/isi.lib.mk"
>>>>>
>>>>>
>>>>>
>>>>> This builds fine as on i386.  I'm trying to get all our user-space to
>>>>> be 64-bit clean, and I run into an error when building on amd64:
>>>>>
>>>>> /data/sb/BR_MDF_64CLEAN/obj/data/sb/BR_MDF_64CLEAN/src/tmp/usr/bin/ld:
>>>>> /data/sb/BR_MDF_64CLEAN/obj/data/sb/BR_MDF_64CLEAN/src/tmp/usr/lib/libl.a(libyywrap.o):
>>>>> relocation R_X86_64_32 can not be used when making a shared object;
>>>>> recompile with -fPIC
>>>>> /data/sb/BR_MDF_64CLEAN/obj/data/sb/BR_MDF_64CLEAN/src/tmp/usr/lib/libl.a:
>>>>> could not read symbols: Bad value
>>>>>
>>>>> The following diff makes the compile work, but I have no idea (yet)
>>>>> whether this will run, if it's the right solution, etc.
>>>>>
>>>>>
>>>>> Index: usr.bin/lex/lib/Makefile
>>>>> ===================================================================
>>>>> --- usr.bin/lex/lib/Makefile    (revision 153343)
>>>>> +++ usr.bin/lex/lib/Makefile    (working copy)
>>>>> @@ -4,11 +4,16 @@
>>>>>
>>>>>  LIB=    ln
>>>>>  SRCS=   libmain.c libyywrap.c
>>>>> -NO_PIC=
>>>>> +#NO_PIC=
>>>>>
>>>>> +SHLIB_MAJOR=   1
>>>>> +SHLIB_MINOR=   0
>>>>> +
>>>>>  .if ${MK_INSTALLLIB} != "no"
>>>>>  LINKS=  ${LIBDIR}/libln.a ${LIBDIR}/libl.a
>>>>>  LINKS+=        ${LIBDIR}/libln.a ${LIBDIR}/libfl.a
>>>>> +LINKS+=        ${LIBDIR}/libln.so ${LIBDIR}/libl.so
>>>>> +LINKS+=        ${LIBDIR}/libln${LIB_SUFFIX}.so ${LIBDIR}/libl${LIB_SUFFIX}.so
>>>>>  .endif
>>>>>
>>>>>  .if ${MK_PROFILE} != "no"
>>>>
>>>> The static-only version was done on purpose:
>>>>
>>>> Revision 1.2: download - view: text, markup, annotated  - select for diffs
>>>> Thu Aug 25 23:11:07 1994 UTC (15 years, 10 months ago) by wollman
>>>> Branches: MAIN
>>>> CVS tags: RELENG_2_1_7_RELEASE, RELENG_2_1_6_RELEASE,
>>>> RELENG_2_1_6_1_RELEASE, RELENG_2_1_5_RELEASE, RELENG_2_1_0_RELEASE,
>>>> RELENG_2_1_0_BP, RELENG_2_0_5_RELEASE, RELENG_2_0_5_BP,
>>>> RELENG_2_0_5_ALPHA, RELENG_2_0_5, RELEASE_2_0, BETA_2_0, ALPHA_2_0
>>>> Branch point for: RELENG_2_1_0
>>>> Diff to: previous 1.1: preferred, colored
>>>> Changes since revision 1.1: +2 -8 lines
>>>>
>>>> We really, really /don't/ want to have a shared lex library.  Also,
>>>> current users should note that the old 1.1.5 lex can't process the
>>>> new scan.l, so you have to copy initscan.c to obj/scan.c before it will
>>>> build.
>>>>
>>>> Garrett Wollman probably has more information about why this was done.
>>>>
>>>> I think that fixing the lib to build with the appropriate options (not
>>>> -m32, or CPUTYPE => some 32-bit x86 variant, etc) is what really needs
>>>> to be done here.
>>
>>> I guess I'm still confused.  The isi_date library compiles fine if
>>> it's for i386, but switching to amd64 gives this error.  Since I
>>> didn't specify any -m32 flags or anything, and it's essentially using
>>> the standard bsd.lib.mk magic, I am trying to figure out why the
>>> 32-bit isi_date.1.so built and the 64-bit one won't.  Was the 32-bit
>>> version building successfully an unfortunate fluke?  What build flags
>>> would get the shared library to link with -ll?
>>
>> I think that amd64 requires a static library be compiled with -fPIC if
>> it's being linked into shared object.  This should not be done for
>> normal static libraries, though, as this could give some performance
>> penalty when it's not needed (i.e. a static binary).
>>
>>> Unfortunately, I didn't write this library, and I don't know anything
>>> about lex(1), so if I need my own yywrap() that might be fine, but I
>>> wouldn't have the first clue what to put in there. :-(
>>
>> I think you could probably just change the code and use %option noyywrap
>> in the .l file?  (do your code call yywrap() directly?)
>
> ^^^^ I mean that the -ll can be just removed for most .l files that have
> noyywrap.

Thanks!  I will try this on Tuesday when I get back to $work,

Cheers,
matthew