[Ohrrpgce] Can't compile 64-bit hspeak on the nightly build vm

James Paige Bob at hamsterrepublic.com
Sun Dec 24 06:45:35 PST 2017 

On Sunday, December 24, 2017, Ralph Versteegen <teeemcee at gmail.com> wrote:

> Hmm, this is quite a mess. I hadn't noticed the -no_pie spelling in
> SConscript. Also some of these are gcc flags and some are linker flags. I
> guess this explains a comment I saw that gcc "standardised on -no-pie".
>
> PIE is only needed on android when building executables, not .so files.
> So, it's only for compiling commandline tests and utilities, not Game
> (which is compiled to libapplication.so)
>
> I feel like keeping builds working on dozens of different platforms and
> configurations and compiler versions is like stumbling around in a dark
> room, and every time you stub your toe on a piece of furniture you try to
> move it out of the way, so never learn where the furniture is!
>
>
That is an apt metaphor!


>
> On 24 December 2017 at 19:00, James Paige <Bob at hamsterrepublic.com> wrote:
>
>> Looking around the SConscript file, I can see that all three of -nopie
>> -no_pie -no-pie seems to exist in one place or another.
>>
>> I also see a comment saying that PIE is mandatory for Android builds.
>>
>> On Sat, Dec 23, 2017 at 9:41 PM, Ralph Versteegen <teeemcee at gmail.com>
>> wrote:
>>
>>> Oh sorry, I got distracted before fixing this.
>>> The way to see whether PIE is the default is apparently to inspect "gcc
>>> -dumpspecs".
>>> The easiest solution is probably to always do a non-PIE build
>>> (otherwise, we have to link the correct FB libraries).
>>> -no-pie is definitely new in GCC 6. What's not clear is whether old
>>> versions always had -nopie. It apparently wasn't documented, at least.
>>>
>>> On 24 December 2017 at 10:45, James Paige <Bob at hamsterrepublic.com>
>>> wrote:
>>>
>>>> So would it be reasonable to have scons run "g++ -no-pie" and parse the
>>>> output to determine whether or not the option is supported?
>>>>
>>>> On a system where it is supported:
>>>>
>>>> james at breq:~$ g++ -no-pie
>>>> g++: fatal error: no input files
>>>> compilation terminated.
>>>>
>>>> On a system where is is not supported:
>>>>
>>>> james at ixrat:~$ g++ -no-pie
>>>> g++: error: unrecognized command line option ‘-no-pie’
>>>> g++: fatal error: no input files
>>>> compilation terminated.
>>>>
>>>>
>>>>
>>>> On Wed, Dec 20, 2017 at 8:41 AM, James Paige <Bob at hamsterrepublic.com>
>>>> wrote:
>>>>
>>>>> Yes, I can confirm that adding "-no-pie" in CXXLINKFLAGS does indeed
>>>>> work. I can just type plain "scons" and it compiles correctly.
>>>>>
>>>>> On Wed, Dec 20, 2017 at 2:35 AM, Ralph Versteegen <teeemcee at gmail.com>
>>>>> wrote:
>>>>>
>>>>>> Bah. I wasn't expecting that to work. fbc apparently only does a PIC
>>>>>> build when building a dynamic library, not when building an executable. I
>>>>>> think what's going on here is that fbc calls ld directly, while linkgcc=1
>>>>>> uses gcc to indirectly call the linker. I guess it's gcc itself which is
>>>>>> telling ld to do a PIE build. I want to fix linkgcc=1.
>>>>>>
>>>>>> So could you please try editing SConscript to replace
>>>>>> CXXLINKFLAGS = []
>>>>>> with
>>>>>> CXXLINKFLAGS = ["-no-pie"]
>>>>>> and see if that works (running just plain "scons")
>>>>>>
>>>>>> On 20 December 2017 at 16:23, James Paige <Bob at hamsterrepublic.com>
>>>>>> wrote:
>>>>>>
>>>>>>> Ah, yes! scons linkgcc=0 did the trick. :)
>>>>>>>
>>>>>>> On Tue, Dec 19, 2017 at 1:10 AM, Ralph Versteegen <
>>>>>>> teeemcee at gmail.com> wrote:
>>>>>>>
>>>>>>>> Hmm. Most of the errors are in libfbmt.a. But a couple of the
>>>>>>>> relocation errors are in .bas modules,  "relocation R_X86_64_32
>>>>>>>> against `.rodata.str1.1' " and I don't understand why - if gcc is
>>>>>>>> defaulting to PIC,
>>>>>>>> As fir why gcc says "...can not be used when making a shared
>>>>>>>> object", I'm guessing that's a gcc bug, since PIC is normally only
>>>>>>>> required when creating a dynamic library on non-x86, so configuring gcc to
>>>>>>>> force it also for executables might activate the same codepath.
>>>>>>>>
>>>>>>>> Anyway, see if any of the following work:
>>>>>>>> scons linkgcc=0
>>>>>>>> FBFLAGS=-pic scons linkgcc=0
>>>>>>>> FBFLAGS="-Wl -no-pie" scons
>>>>>>>> (I suggest linkgcc=0 because linkgcc=1 always links libfbmt.a,
>>>>>>>> never libfbmtpic.a which is compiled with -fPIC)
>>>>>>>>
>>>>>>>> On 19 December 2017 at 16:47, James Paige <Bob at hamsterrepublic.com>
>>>>>>>> wrote:
>>>>>>>>
>>>>>>>>> I should have mentioned I already tried scons -c and it didn't
>>>>>>>>> help :(
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> On Monday, December 18, 2017, Ralph Versteegen <teeemcee at gmail.com>
>>>>>>>>> wrote:
>>>>>>>>>
>>>>>>>>>> Run "scons -c" to delete the old object files. Since they were
>>>>>>>>>> compiled without -fPIC, which is now the default in debian 9, they can't be
>>>>>>>>>> linked into a PIE executable.
>>>>>>>>>>
>>>>>>>>>> On 19 December 2017 at 14:05, James Paige <
>>>>>>>>>> Bob at hamsterrepublic.com> wrote:
>>>>>>>>>>
>>>>>>>>>>> So I made a copy of the 64 bit Linux VM, and upgraded it from
>>>>>>>>>>> Debian 8 to Debian 9.
>>>>>>>>>>>
>>>>>>>>>>> hspeak compiles fine now as a 64 bit binary, but I can no longer
>>>>>>>>>>> compile ohrrpgce-game and ohrrpgce-custom
>>>>>>>>>>>
>>>>>>>>>>> when making a shared object; recompile with -fPIC
>>>>>>>>>>> /usr/bin/ld: /usr/local/bin/../lib/freebasic/linux-x86_64/libfbmt.a(io_print_wstr.o): relocation R_X86_64_32 against undefined symbol `__fb_ctx' can not be used when making a shared object; recompile with -fPIC
>>>>>>>>>>> /usr/bin/ld: /usr/local/bin/../lib/freebasic/linux-x86_64/libfbmt.a(io_printvoid_wstr.o): relocation R_X86_64_32 against `.rodata.str4.4' can not be used when making a shared object; recompile with -fPIC
>>>>>>>>>>> /usr/bin/ld: /usr/local/bin/../lib/freebasic/linux-x86_64/libfbmt.a(sys_getexename.o): relocation R_X86_64_32 against `.rodata.str1.1' can not be used when making a shared object; recompile with -fPIC
>>>>>>>>>>> /usr/bin/ld: /usr/local/bin/../lib/freebasic/linux-x86_64/libfbmt.a(file_print_wstr.o): relocation R_X86_64_32 against undefined symbol `__fb_ctx' can not be used when making a shared object; recompile with -fPIC
>>>>>>>>>>> /usr/bin/ld: /usr/local/bin/../lib/freebasic/linux-x86_64/libfbmt.a(io_printpad_wstr.o): relocation R_X86_64_32 against `.rodata.str4.4' can not be used when making a shared object; recompile with -fPIC
>>>>>>>>>>> /usr/bin/ld: /usr/local/bin/../lib/freebasic/linux-x86_64/libfbmt.a(dev_scrn_write_wstr.o): relocation R_X86_64_32S against symbol `fb_DevScrnWriteWstr' can not be used when making a shared object; recompile with -fPIC
>>>>>>>>>>> /usr/bin/ld: /usr/local/bin/../lib/freebasic/linux-x86_64/libfbmt.a(io_printbuff_wstr.o): relocation R_X86_64_32 against `.rodata.str1.1' can not be used when making a shared object; recompile with -fPIC
>>>>>>>>>>> /usr/bin/ld: /usr/local/bin/../lib/freebasic/linux-x86_64/libfbmt.a(array_erase.o): relocation R_X86_64_PC32 against symbol `free@@GLIBC_2.2.5' can not be used when making a shared object; recompile with -fPIC
>>>>>>>>>>> /usr/bin/ld: final link failed: Bad value
>>>>>>>>>>> collect2: error: ld returned 1 exit status
>>>>>>>>>>> scons: *** [ohrrpgce-game] Error 1
>>>>>>>>>>> scons: building terminated because of errors.
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> On Sun, Dec 17, 2017 at 10:21 AM, James Paige <
>>>>>>>>>>> Bob at hamsterrepublic.com> wrote:
>>>>>>>>>>>
>>>>>>>>>>>> Today I will try upgrading Debian on a copy of the 64-bit VM,
>>>>>>>>>>>>  Hopefully that will resolve it.
>>>>>>>>>>>>
>>>>>>>>>>>> ---
>>>>>>>>>>>> James
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> On Sat, Dec 16, 2017 at 4:00 PM, Ralph Versteegen <
>>>>>>>>>>>> teeemcee at gmail.com> wrote:
>>>>>>>>>>>>
>>>>>>>>>>>>> -no-pie is actually passed by SConscript.  I can't find any
>>>>>>>>>>>>> mention in my emails about a -no-pie problem on debian. The problem we had
>>>>>>>>>>>>> was that -no-pie was supported (nor necessary) for clang on OSX.
>>>>>>>>>>>>>
>>>>>>>>>>>>> The -no-pie flag was added in r9542:
>>>>>>>>>>>>> """
>>>>>>>>>>>>> scons: Fix hspeak builds on Arch Linux and others where gcc
>>>>>>>>>>>>> defaults to PIE
>>>>>>>>>>>>>
>>>>>>>>>>>>> The Euphoria 4.1 builds we recommend (and which are also used
>>>>>>>>>>>>> by the Arch
>>>>>>>>>>>>> opeeuphoria package on AUR) haven't been built for PIE
>>>>>>>>>>>>> """
>>>>>>>>>>>>> Apparently this flag is also necessary on (recent?) Debian and
>>>>>>>>>>>>> Ubuntu, as they also default to PIE.
>>>>>>>>>>>>>
>>>>>>>>>>>>> It's not clear, but from searching around it appears that this
>>>>>>>>>>>>> flag used to be called -nopie until gcc 6 switched to -no-pie. So try that.
>>>>>>>>>>>>> I have the man page for gcc 4.8 here, and it doesn't say either way.
>>>>>>>>>>>>> Also, there is -fno-pie which is a compiler flag while -no-pie
>>>>>>>>>>>>> is a linker flag (also understood by gcc). So you could also try that and
>>>>>>>>>>>>> see whether it works. (I assume that -no-pie passed to gcc implies -fno-pie)
>>>>>>>>>>>>>
>>>>>>>>>>>>> On 16 December 2017 at 12:3
>>>>>>>>>>>>> 2, James Paige <Bob at hamsterrepublic.com> wrote:
>>>>>>>>>>>>>
>>>>>>>>>>>>>> So now I remember why I have the 32-bit version of Euphoria
>>>>>>>>>>>>>> installed on the Linux nightly build machine.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> The hspeak build fails because gcc doesn't understand --no-pie
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> gcc -o /home/james/src/ohr/wip/hspeak init-.o hspeak.o
>>>>>>>>>>>>>> main-.o hspeak_0.o hspeak_1.o hspeak_2.o hspeak_3.o hspeak_4.o hsspiffy.o
>>>>>>>>>>>>>> file.o sort.o get.o graphics.o machine.o math.o types.o unicode.o
>>>>>>>>>>>>>> ucstypes.o io_with_unicode.o 0achine.o memory.o dll.o
>>>>>>>>>>>>>> /usr/share/euphoria/bin/eu.a -m64 -ldl -lm -lpthread -no-pie
>>>>>>>>>>>>>> gcc: error: unrecognized command line option '-no-pie'
>>>>>>>>>>>>>> hspeak.mak:97: recipe for target
>>>>>>>>>>>>>> '/home/james/src/ohr/wip/hspeak' failed
>>>>>>>>>>>>>> make: *** [/home/james/src/ohr/wip/hspeak] Error 1
>>>>>>>>>>>>>> make: Leaving directory '/home/james/src/ohr/wip/build
>>>>>>>>>>>>>> /hspeak'
>>>>>>>>>>>>>> scons: done building targets.
>>>>>>>>>>>>>> james at ixrat:~/src/ohr/wip$ euc -v
>>>>>>>>>>>>>> Euphoria to C Translator v4.1.0 development
>>>>>>>>>>>>>>    64-bit Linux, Using System Memory
>>>>>>>>>>>>>>    Revision Date: 2015-02-02 14:18:53, Id: 5861:57179171dbed
>>>>>>>>>>>>>> james at ixrat:~/src/ohr/wip$ gcc --version
>>>>>>>>>>>>>> gcc (Debian 4.9.2-10) 4.9.2
>>>>>>>>>>>>>> Copyright (C) 2014 Free Software Foundation, Inc.
>>>>>>>>>>>>>> This is free software; see the source for copying
>>>>>>>>>>>>>> conditions.  There is NO
>>>>>>>>>>>>>> warranty; not even for MERCHANTABILITY or FITNESS FOR A
>>>>>>>>>>>>>> PARTICULAR PURPOSE.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> I am wondering if the easiest way to resolve this might be to
>>>>>>>>>>>>>> upgrade the virtual machine from Debian 8.x to 9.x and hopefully get a
>>>>>>>>>>>>>> version of gcc that does not have this problem.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> For now, I am leaving the 64-bit version of euphoria
>>>>>>>>>>>>>> installed on the vm, so the 64-bit nightly builds are going to remain
>>>>>>>>>>>>>> broken until I have figured out a solution.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> ---
>>>>>>>>>>>>>> James
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> _______________________________________________
>>>>>>>>>>>>>> Ohrrpgce mailing list
>>>>>>>>>>>>>> ohrrpgce at lists.motherhamster.org
>>>>>>>>>>>>>> http://lists.motherhamster.org/listinfo.cgi/ohrrpgce-motherh
>>>>>>>>>>>>>> amster.org
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>> _______________________________________________
>>>>>>>>>>>>> Ohrrpgce mailing list
>>>>>>>>>>>>> ohrrpgce at lists.motherhamster.org
>>>>>>>>>>>>> http://lists.motherhamster.org/listinfo.cgi/ohrrpgce-motherh
>>>>>>>>>>>>> amster.org
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> _______________________________________________
>>>>>>>>>>> Ohrrpgce mailing list
>>>>>>>>>>> ohrrpgce at lists.motherhamster.org
>>>>>>>>>>> http://lists.motherhamster.org/listinfo.cgi/ohrrpgce-motherh
>>>>>>>>>>> amster.org
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>> _______________________________________________
>>>>>>>>> Ohrrpgce mailing list
>>>>>>>>> ohrrpgce at lists.motherhamster.org
>>>>>>>>> http://lists.motherhamster.org/listinfo.cgi/ohrrpgce-motherh
>>>>>>>>> amster.org
>>>>>>>>>
>>>>>>>>>
>>>>>>>>
>>>>>>>> _______________________________________________
>>>>>>>> Ohrrpgce mailing list
>>>>>>>> ohrrpgce at lists.motherhamster.org
>>>>>>>> http://lists.motherhamster.org/listinfo.cgi/ohrrpgce-motherh
>>>>>>>> amster.org
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> Ohrrpgce mailing list
>>>>>>> ohrrpgce at lists.motherhamster.org
>>>>>>> http://lists.motherhamster.org/listinfo.cgi/ohrrpgce-motherh
>>>>>>> amster.org
>>>>>>>
>>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> Ohrrpgce mailing list
>>>>>> ohrrpgce at lists.motherhamster.org
>>>>>> http://lists.motherhamster.org/listinfo.cgi/ohrrpgce-motherh
>>>>>> amster.org
>>>>>>
>>>>>>
>>>>>
>>>>
>>>> _______________________________________________
>>>> Ohrrpgce mailing list
>>>> ohrrpgce at lists.motherhamster.org
>>>> http://lists.motherhamster.org/listinfo.cgi/ohrrpgce-motherhamster.org
>>>>
>>>>
>>>
>>> _______________________________________________
>>> Ohrrpgce mailing list
>>> ohrrpgce at lists.motherhamster.org
>>> http://lists.motherhamster.org/listinfo.cgi/ohrrpgce-motherhamster.org
>>>
>>>
>>
>> _______________________________________________
>> Ohrrpgce mailing list
>> ohrrpgce at lists.motherhamster.org
>> http://lists.motherhamster.org/listinfo.cgi/ohrrpgce-motherhamster.org
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.motherhamster.org/pipermail/ohrrpgce-motherhamster.org/attachments/20171224/c77b469d/attachment-0001.html>

More information about the Ohrrpgce mailing list