CSV Read Fails on dtypes.int32 After Recent int_ Removal

[stanbrub]

The recent removal of dtypes.int_ caused an error in a benchmark. When switching to dtypes.int32 instead, the benchmark still fails but with a different error. I've attached the stack trace.

It looks like the CSV read does not support all the scalar dtypes. I'm looking for a true 32 bit int, not a 64 bit one, for my test. (Though I'm not sure that's doable from the python side. Maybe np.int32?) Exception is raised from /opt/deephaven/venv/lib/python3.10/site-packages/deephaven/csv.py.

csv-read-datatype-exception.txt

[devinrsmith]

dtypes.int_ was dtypes.int64 not dtypes.int32; try with dtypes.int64 to see if it works.

[stanbrub]

Here's an example of using int32 in the console that fails.

from deephaven import dtypes as dht, empty_table
from deephaven.csv import read, write

source = empty_table(10).update(['col1=i'])
write(source, '/data/dtypes.bug.csv')
result = read('/data/dtypes.bug.csv',{'int32':dht.int32})

If I read with int64, the result column is a 64 bit int. With int32, it fails with exception. With no types specified, the result column is a 32 bit int.

[devinrsmith]

I noticed that there is a mis-mapping to parsers in csv.py, but it seems like there is a more fundamental issue (even after fixing the mappings):

from deephaven import read_csv, dtypes

actual = read_csv('primitive_types.csv', {
    'bool_': dtypes.bool_,
    'byte': dtypes.byte,
    'char': dtypes.char,
    'short': dtypes.short,
    'int32': dtypes.int32,
    'long': dtypes.long,
    'float32': dtypes.float32,
    'double': dtypes.double,
})

bool_,byte,char,short,int32,long,float32,double
true,42,a,42,42,42,42.42,42.42

The specs look correct:

The results wrong:

May need to file issue in https://github.com/deephaven/deephaven-csv