An warning about ccall occurs

[zhujinxuan]

After upgrading julia from 0.3.0-prerelease+91 to 0.3.0-prerelease+618 , an error occurs when I'm reading NetCDF with NetCDF.jl:

julia> ncinfo("HadISST_sst.nc")
 warning: literal address used in ccall for (null); code cannot be statically compiled 
 warning: literal address used in ccall for (null); code cannot be statically compiled 
 warning: literal address used in ccall for (null); code cannot be statically compiled 

Does this warn me that julia have not been compiled correctly in my computer?

[timholy]

No, there's no major problem here. To eliminate the warning, the NetCDF.jl package should be modified to use a string rather than a pointer for the library in the ccall. I'd recommend filing an issue with the package.

To the rest: I'd close this, but it prompts a thought: should we have a documentation page that expands on the meaning of particular errors? Julia's error messages are brief and effective, but employ a fair amount of assumed knowledge/jargon:

• "What's a literal address?"
• "What does 'statically compiled' mean?"
• "Wait, I thought Julia was JIT-compiled?!"
• "I didn't write a ccall to (null)."
• "I'm not trying to statically-compile my package"

etc.

[JeffBezanson]

We should probably not show that error outside of "imaging mode".

[ivarne]

@timholy Usually google does a very good job of finding a discussion about what the root cause of a error is in a particular instance, but if Julia erros/exceptions could include a link to a page on julialang.org with some discussion that would be wonderful. The trouble is that it needs to be kept up to date.

[timholy]

@ivarne, good point.

@JeffBezanson, agreed. Out of curiosity, when not in "imaging mode", is there a performance difference between pointer-ccalls and string-ccalls? Several libraries have recently switched (due to this warning).

[JeffBezanson]

I don't think there is a performance difference, but you should avoid
calling direct pointers when possible.
On Dec 22, 2013 1:10 PM, "Tim Holy" [email protected] wrote:

@ivarne https://github.com/ivarne, good point.

@JeffBezanson https://github.com/JeffBezanson, agreed. Out of
curiosity, when not in "imaging mode", is there a performance difference
between pointer-ccalls and string-ccalls? Several libraries have recently
switched (due to this warning).

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/5215#issuecomment-31092407
.

[vtjnash]

Pointer-ccalls that got this warning would segfault if compiled into the system image. I added this warning specifically for the non-imaging mode case so that packages code would get this deprecation warning, especially since they can already be added to the sysimg. Can we instead improve the error message & add a link to some added documentation to help the user? (actually, can we just add error codes to all of our error messages and write a cross-reference page with extended documentation describing the error -- #6386)