Module loading issue related to differences in dlopen default flags in macOS and linux

[npm-sdr]

Module loading issue related to differences in dlopen default flags in macOS and linux

[zuckschwerdt]

Out of curiosity, which driver modules did you encounter this problem with?

[npm-sdr]

It's a custom driver module I wrote for my own hardware - I started with rtlsdr Soapy driver module as a starting point. I have two different generations of H/W and for this I created two different Soapy driver modules so that I can link each Soapy driver module against a hardware specific low-level driver library. I've given these Soapy driver modules unique driver names as below and built 2 independent shared libraries:

static SoapySDR::Registry registerSDRFE("soapy_sdrfe_sc5", &findSDRFE, &makeSDRFE, SOAPY_SDR_ABI_VERSION);

and

static SoapySDR::Registry registerSDRFE("soapy_sdrfe_sc7", &findSDRFE, &makeSDRFE, SOAPY_SDR_ABI_VERSION);

[zuckschwerdt]

Effectively all your symbols are duplicated, and perhaps one of registerSDRFE, findSDRFE, makeSDRFE then creates this problem? You might be the first one to encounter this oddity then :) It should go away when stripping the symbols (nm -u) from the modules, right?

Nevertheless the suggestion to explicity choose RTLD_LOCAL seems a good idea.

[npm-sdr]

I tried renaming findSDRFE/makeSDRFE to findSDRFE_sc5/makeSDRFE_sc5 and findSDRFE_sc7/makeSDRFE_sc7 in the 2 Soapy driver modules and that didn't help. Also tried stripping the symbols and that didn't work either.

[npm-sdr]

@zuckschwerdt, is SoapySDR still being maintained? There hasn't been any commits or bug fixes since middle of last year.

[zuckschwerdt]

There are admins still around and active maintainers tending mostly to the modules. I guess development slowed down as things are working well and are somewhat feature-complete?

I'll PR this fix and merge it in ~2 weeks -- should give everyone a chance to chime in if needed.

[npm-sdr]

There are admins still around and active maintainers tending mostly to the modules. I guess development slowed down as things are working well and are somewhat feature-complete?

I'll PR this fix and merge it in ~2 weeks -- should give everyone a chance to chime in if needed.

Great - thank you.