Follow up changes for InitError #12742
Conversation
| ccall((:MKL_Set_Interface_Layer, Base.libblas_name), Void, (Cint,), USE_BLAS64 ? 1 : 0) | ||
| end | ||
| catch ex | ||
| #Base.showerror_nostdio(ex, |
There was a problem hiding this comment.
was this not able to show the details from the exception?
There was a problem hiding this comment.
No, I just mistakenly didn't revert some of my last tests. Fixing now.
jdlangs
force-pushed
the
init_exceptions
branch
from
13dacd3 to
419539d
Compare
|
Hmm, needing to wrap all of these |
| end | ||
| try | ||
| if gmp_version().major != GMP_VERSION.major || gmp_bits_per_limb() != GMP_BITS_PER_LIMB | ||
| error(string("The dynamically loaded GMP library (version $(gmp_version()) with __gmp_bits_per_limb == $(gmp_bits_per_limb()))\n", |
There was a problem hiding this comment.
Shouldn't this just be warn(string(.... The try... catch seems strange, when the only possible exception comes from an explicit error()
Please verify if warn work in __init__, there might be issues with STDOUT there too.
There was a problem hiding this comment.
It's not just that error call, but any ccall into libgmp that might throw. The goal is to let Base load even when any of the shared libraries are actually missing, though I don't know how that situation might arise.
warn was the first thing I tried and it didn't work which led to the new showerror function.
There was a problem hiding this comment.
though I don't know how that situation might arise.
people embedding julia into a different language or wanting to deploy standalone executables might not always need bigint's
|
It should provide good incentive to not do shared library loading in Base if possible. Another option would be to catch all |
|
Bump. Is this ready? |
|
It would be nice to leave a note somewhere explaining why the |
| end | ||
| catch ex | ||
| Base.showerror_nostdio(ex, | ||
| "WARNING: Error during initialization of module LinAlg") |
There was a problem hiding this comment.
Missing the :\n at the end of this message. However I think it's better to print the formatting stuff (colon and newline) where it's needed rather than making it part of the message.
There was a problem hiding this comment.
Moving it to the showerror function now.
Prevents them from throwing any InitError exceptions that will not get handled
jdlangs
force-pushed
the
init_exceptions
branch
from
419539d to
fe0337e
Compare
Follow up changes for InitError
Per the comments in #12576, this no longer escapes
InitErrors from being wrapped inLoadErrors.I also wrapped all the Base submodule
__init__functions that use external libraries intry...catchso they don't throw but I'm not sure if it's the best approach to take.