[Scons-users] scons-local-2.5.1 build and clean shows random failure
Hua Yanghao
huayanghao at gmail.com
Thu Jul 27 08:17:52 EDT 2017
Thanks Bill.
Your engineering attitude really makes me feeling
a lot more confident using scons :-)
If VariantDir() does not allow "." as src direction
while output is within "." (e.g. "./output") then VariantDir()
should probably check for this condition and prevent
user from having this setup.
Together if you could manage to reproduce the unreliable
build clean so far I don't see any blocker to use scons
and pushing it to a wider audience.
Thanks a lot for your support!
On Thu, Jul 27, 2017 at 4:52 AM, Bill Deegan <bill at baddogconsulting.com> wrote:
> I'm not sure why it didn't consistently fail.
> I'll try to make a simple testcase and file a bug.
>
> Glad that you've resolved your issue for now.
>
> -Bill
>
> On Wed, Jul 26, 2017 at 6:22 PM, Bill Deegan <bill at baddogconsulting.com>
> wrote:
>>
>> See comments inline below..
>>
>>
>> On Wed, Jul 26, 2017 at 11:55 AM, Hua Yanghao <huayanghao at gmail.com>
>> wrote:
>>>
>>> I see. So the correct use of VariantDir() is to
>>> restrict the second parameter, e.g. the src to
>>> be something else rather than "."?
>>>
>>> e.g. VariantDir() can only take the form:
>>> VairantDir(build_dir, src_dir) where src_dir cannot
>>> be ".".
>>>
>>> Worth to add a check that VariantDir() cannot take "."
>>> as its src directory?
>>
>>
>> Having . as src is fine as long as the variantdir is not under .
>> So:
>> VariantDir('../build/linux64','.')
>> Would work fine.
>>
>>>
>>>
>>> Consider my layout of source code is like this:
>>> .
>>> ├── arch
>>> ├── boards
>>> ├── common
>>> ├── configs
>>> ├── doc
>>> ├── drivers
>>> ├── external
>>> ├── include
>>> ├── lib
>>> ├── Makefile
>>> ├── output
>>> ├── README.md
>>> ├── SConstruct
>>> ├── site_scons
>>> ├── soc
>>> ├── test
>>> ├── TODO
>>> └── tools
>>>
>>> And besides "output" all other folders can be a src folder.
>>> And I thought I can use VariantDir() pointing to "output/linux64_full"
>>> for all SConscript that SConstruct includes.
>>
>>
>> A more typical scons layout would be
>> src/<arch,boards,... all your source dirs>
>> output/<all your variants>
>> ... rest of not source directories.
>> Then VariantDir('output/linux64','src')
>> would work fine.
>>
>>
>>>
>>>
>>> On Wed, Jul 26, 2017 at 5:15 PM, Bill Deegan <bill at baddogconsulting.com>
>>> wrote:
>>> > There is not general issue with VariantDir(), just the way you used it
>>> > created a dependency cycle.
>>> >
>>> >
>>> > output/a was created from .
>>> > Then next run
>>> > output/a/output/a was created from . into output/a
>>> > Now output/a/output/a depends on output/a... see the loop?
>>> >
>>> > On Wed, Jul 26, 2017 at 4:53 AM, Hua Yanghao <huayanghao at gmail.com>
>>> > wrote:
>>> >>
>>> >> Bill, it works now!
>>> >>
>>> >> I overlooked one thing the SConscript variant_dir should point to
>>> >> its specific output dir not the global one for all SConscript file.
>>> >>
>>> >> + variant_dir = os.path.join(output, folder)
>>> >> + lib, objs, objs_all = ENV.SConscript(true_scons_file,
>>> >> variant_dir=variant_dir)
>>> >>
>>> >> This this change I am able to build successfully. The other changes
>>> >> that required is
>>> >> for the include path you need to specific the "true include" path not
>>> >> the one that is
>>> >> going to be copied into the output/linux64_full folder. With this two
>>> >> changes everything
>>> >> works fine with 100 build clean -j8 not a single time reproduced any
>>> >> issue.
>>> >>
>>> >> Even though I think I still don't fully understood the reason why
>>> >> VariantDir() caused
>>> >> the issue. should VariantDir() be removed/deprecated or should it be
>>> >> changed in a way
>>> >> that it has the same behavior of variant_dir parameter to
>>> >> env.SConscript()?
>>> >>
>>> >> Thanks a lot for your help.
>>> >>
>>> >> On Wed, Jul 26, 2017 at 12:22 PM, Hua Yanghao <huayanghao at gmail.com>
>>> >> wrote:
>>> >> > I tried first with the SConscript variant_dir parameter but seems it
>>> >> > doesn't work:
>>> >> >
>>> >> > + lib, objs, objs_all = ENV.SConscript(true_scons_file,
>>> >> > variant_dir=output)
>>> >> >
>>> >> > And the result is:
>>> >> > hua at grass:~/git/usw $ configs/linux64_full.py
>>> >> > Running : ./tools/scons config=linux64_full.py
>>> >> > scons: Reading SConscript files ...
>>> >> > output dir: output/linux64_full
>>> >> > output dir: output/linux64_full
>>> >> >
>>> >> > scons: *** 'output/linux64_full' already has a source directory:
>>> >> > 'arch/linux64_full'.
>>> >> > File "/home/hua/git/usw/site_scons/common.py", line 61, in
>>> >> > BuildObjList
>>> >> > hua at grass:~/git/usw $
>>> >> >
>>> >> > I found that if I want the the build to happen in the
>>> >> > output/linux64_full folder
>>> >> > I have to call SConscript file from e.g.
>>> >> > output/linux64_full/src/SConscript instead of
>>> >> > the top level true SConscript file (src/SConscript).
>>> >> >
>>> >> > Did I use it wrong?
>>> >> >
>>> >> > On Wed, Jul 26, 2017 at 12:39 AM, Bill Deegan
>>> >> > <bill at baddogconsulting.com> wrote:
>>> >> >> Yes.. You created dependency cycle..
>>> >> >> But only after a build.
>>> >> >> Thus why it's showing up only on clean...
>>> >> >>
>>> >> >> Once you build you have output/linux64_build
>>> >> >> Then you say map . to output/linux_64build which ends up with a
>>> >> >> virtual
>>> >> >> subdirectory via variantdir of
>>> >> >> output/linux64_build/output/linux_64...
>>> >> >>
>>> >> >> see the issue.
>>> >> >>
>>> >> >> Variantdirs should never have parent/child relationships.
>>> >> >>
>>> >> >> I'd advice instead of using VariantDir(), use variant_dir in your
>>> >> >> SConscript
>>> >> >> call to map like this:
>>> >> >>
>>> >> >> SConscript('src/SConscript',variant_dir='output/linux64_build/src')
>>> >> >> etc..
>>> >> >>
>>> >> >> See if doing that resolves the issue?
>>> >> >>
>>> >> >> On Tue, Jul 25, 2017 at 1:03 PM, Hua Yanghao <huayanghao at gmail.com>
>>> >> >> wrote:
>>> >> >>>
>>> >> >>> Yes, that's exactly how I used VariantDir. Any problem seen?
>>> >> >>>
>>> >> >>> On Tue, Jul 25, 2017 at 6:32 PM, Bill Deegan
>>> >> >>> <bill at baddogconsulting.com>
>>> >> >>> wrote:
>>> >> >>> > Can you paste just the usage of VariantDir() from your build?
>>> >> >>> >
>>> >> >>> > Are you doing
>>> >> >>> >
>>> >> >>> > VariantDir('output/linux64_full','.') in your SConstruct?
>>> >> >>> >
>>> >> >>> >
>>> >> >>> > On Tue, Jul 25, 2017 at 8:47 AM, Hua Yanghao
>>> >> >>> > <huayanghao at gmail.com>
>>> >> >>> > wrote:
>>> >> >>> >>
>>> >> >>> >> I think I am in, can you see my message there? Got no response.
>>> >> >>> >>
>>> >> >>> >> On Tue, Jul 25, 2017 at 5:45 PM, Bill Deegan
>>> >> >>> >> <bill at baddogconsulting.com>
>>> >> >>> >> wrote:
>>> >> >>> >> > goto:
>>> >> >>> >> > http://webchat.freenode.net/
>>> >> >>> >> >
>>> >> >>> >> > pick any nickname
>>> >> >>> >> > channel is "#scons"
>>> >> >>> >> >
>>> >> >>> >> > On Tue, Jul 25, 2017 at 8:43 AM, Hua Yanghao
>>> >> >>> >> > <huayanghao at gmail.com>
>>> >> >>> >> > wrote:
>>> >> >>> >> >>
>>> >> >>> >> >> On Tue, Jul 25, 2017 at 5:42 PM, Bill Deegan
>>> >> >>> >> >> <bill at baddogconsulting.com>
>>> >> >>> >> >> wrote:
>>> >> >>> >> >> > so you see this most often when:
>>> >> >>> >> >> >
>>> >> >>> >> >> > scons -j <SOME NUMBER> -c ?
>>> >> >>> >> >>
>>> >> >>> >> >> Exactly, it fails most often with scons -j8 -c, very rare
>>> >> >>> >> >> with
>>> >> >>> >> >> scons
>>> >> >>> >> >> -j8.
>>> >> >>> >> >>
>>> >> >>> >> >> > Any chance you can hop on the IRC channel?
>>> >> >>> >> >> >
>>> >> >>> >> >>
>>> >> >>> >> >> If you give me an instruction for howto ... I could try,
>>> >> >>> >> >> never
>>> >> >>> >> >> really
>>> >> >>> >> >> used
>>> >> >>> >> >> IRC.
>>> >> >>> >> >> _______________________________________________
>>> >> >>> >> >> Scons-users mailing list
>>> >> >>> >> >> Scons-users at scons.org
>>> >> >>> >> >> https://pairlist4.pair.net/mailman/listinfo/scons-users
>>> >> >>> >> >
>>> >> >>> >> >
>>> >> >>> >> >
>>> >> >>> >> > _______________________________________________
>>> >> >>> >> > Scons-users mailing list
>>> >> >>> >> > Scons-users at scons.org
>>> >> >>> >> > https://pairlist4.pair.net/mailman/listinfo/scons-users
>>> >> >>> >> >
>>> >> >>> >> _______________________________________________
>>> >> >>> >> Scons-users mailing list
>>> >> >>> >> Scons-users at scons.org
>>> >> >>> >> https://pairlist4.pair.net/mailman/listinfo/scons-users
>>> >> >>> >
>>> >> >>> >
>>> >> >>> >
>>> >> >>> > _______________________________________________
>>> >> >>> > Scons-users mailing list
>>> >> >>> > Scons-users at scons.org
>>> >> >>> > https://pairlist4.pair.net/mailman/listinfo/scons-users
>>> >> >>> >
>>> >> >>> _______________________________________________
>>> >> >>> Scons-users mailing list
>>> >> >>> Scons-users at scons.org
>>> >> >>> https://pairlist4.pair.net/mailman/listinfo/scons-users
>>> >> >>
>>> >> >>
>>> >> >>
>>> >> >> _______________________________________________
>>> >> >> Scons-users mailing list
>>> >> >> Scons-users at scons.org
>>> >> >> https://pairlist4.pair.net/mailman/listinfo/scons-users
>>> >> >>
>>> >> _______________________________________________
>>> >> Scons-users mailing list
>>> >> Scons-users at scons.org
>>> >> https://pairlist4.pair.net/mailman/listinfo/scons-users
>>> >
>>> >
>>> >
>>> > _______________________________________________
>>> > Scons-users mailing list
>>> > Scons-users at scons.org
>>> > https://pairlist4.pair.net/mailman/listinfo/scons-users
>>> >
>>> _______________________________________________
>>> Scons-users mailing list
>>> Scons-users at scons.org
>>> https://pairlist4.pair.net/mailman/listinfo/scons-users
>>
>>
>
>
> _______________________________________________
> Scons-users mailing list
> Scons-users at scons.org
> https://pairlist4.pair.net/mailman/listinfo/scons-users
>
More information about the Scons-users
mailing list