Re: [Boost-testing] CrystaX runners

Subject: Re: [Boost-testing] CrystaX runners
From: Raffi Enficiaud (raffi.enficiaud_at_[hidden])
Date: 2015-03-20 12:18:38

Le 20/03/15 17:03, Dmitry Moskalchuk a écrit :
>
>> I believe this scheme cannot detect at all any relative path that is
>> encoded in the binary itself, which is the problem in our case.
>> The binary implicitly supposes that it is running inside the same
>> folder as in the one of the source tree.
>
> Yes, it can't, but we can fix adbrunner itself, adding specific code
> branch so it know that in case of executable named
> "error_handling_test", it should additionally copy hard-coded set of
> files to device. This is not ideal, I know, but really I don't see
> another way how to fix that without changing of test code itself.
>
> Alternatively, you can fix sources of test and pass directory with files
> as a command line argument (I personally would prefer this way since
> it's less error prone).
>
>> Wouldn't it be possible to mount a volume to those devices instead of
>> copying? Or to copy the full source tree (maybe without the build
>> folder)?
>
> Unfortunately, it's just impossible to mount any folder to Android
> device - 'adb' can't do that and network mounting just don't work.
> Copying of whole sources tree most likely will not work, just because
> it's too big for Android devices - as I see, right now whole Boost
> source tree (without .git folder) is ~380 Mb, which is really too big
> for Android devices or emulators (usually Android devices don't have so
> much free space available for executables + their data). It depends on
> specific device, but I clearly see that even though it might work on
> some hi-end devices, many other (especially low-end) devices would just
> be excluded from testing at all. I can try to do that though and see how
> it will be really, but don't expect results too quickly (not earlier
> than next week).
>
> Anyway, this is not very good solution (IMHO) since this is not
> Android-specific problem. The same would be for any other cross-platform
> builds if it would be added in the future. So I really think it would be
> better to pass paths to files/directories as a command-line arguments to
> specific tests.
>

For cross platform builds, I usually mount the build path on the virtual 
machine. I really do not know anything about Android.

Having now a closer look at the Jamfile, it appears that the files are 
passed on the command line. So, according to what you are saying, "it 
should" work. These paths are relative, so it looks like bjam is doing 
some magic there (running from the test folder maybe).

Best,
Raffi