basic support for try except

[sklam]

Implements #4898 (comment)

[sklam]

• f127335 fixes basic support for try except #4902 (comment) and it's not an elegant way but will do the job before all the old-style impl is updated
• 6af3ef1 fixes basic support for try except #4902 (comment)

[seibert]

now has conflict with master

[stuartarchibald]

Thanks for the fix-ups, couple of typo's else good to go.

[stuartarchibald]

Thanks for your work on this, great to see this feature added!

[esc]

🎉

[mobassir94]

i tried the code above :

@njit
def udt():
    x = 1
    y = 0
    try:
        x / y
    except Exception:
        pass

udt()

in numba version 0.47 and 0.48 and it still throws me error : "UnsupportedError: Failed in nopython mode pipeline (step: analyzing bytecode)
'try' block not supported until python3.7 or later"

how to solve this issue?

[esc]

@mobassir94 I just tried the snippet you suggested on the current Numba master 0.50.0dev0-7-g917f582433-dirty and it seems to be fine. Could you give it a try?

[mobassir94]

@esc which one?
i tried this :

@njit
def udt():
    x = 1
    y = 0
    try:
        x / y
    except Exception:
        pass

udt()

and it didn't work at all in kaggle kernel

another code that i was trying is this one :

%%time
from langdetect import detect
count = 0
noise = []
non_eng = []
for i in range(len(df_train_full)):
    try:
        lang = detect(df_train_full.comment_text[i])
        if(lang != 'en'):
            count +=1
            non_eng.append(i)
    except:
        noise.append(i)

and the error comes for this line of code : lang = detect(df_train_full.comment_text[i])
i can't use this library : https://pypi.org/project/langdetect/ in try block,will you please give it a try and see if you can make it work?
the dataset i was using is this one : https://www.kaggle.com/c/jigsaw-multilingual-toxic-comment-classification/data

[esc]

@mobassir94 thanks for following up. We appreciate your input in this matter. I am not sure what you mean "it didn't work at all in kaggle kernel"?! I just tried it again locally:

$ cat foo033.py
from numba import njit


@njit
def udt():
    x = 1
    y = 0
    try:
        x / y
    except Exception:
        pass


udt()
$ python foo033.py
$ echo $?
0
$ git desc
0.50.0dev0-7-g917f582433

As you can see I am not receiving any error message. Perhaps you could submit the full backtrace of the exception? Also, what Python version are you using? As can be seen on https://pypi.org/project/numba/ Numba only supports Pythonn 3.6 and later.

As for the issues with langdetect, only a subset of NumPy and Python functions/features are supported by Numba, documentation on those supported can be found here.

• NumPy Features
• Python Features

Effectively this means you can not use langdetect in functions that you would like to compile with Numba.

[gmarkall]

@mobassir94 Could you please open an issue for further discussion please? It will also be helpful for you to report the numba version used in the Kaggle kernel, i.e. numba.__version__ - many thanks!

[mobassir94]

dear @gmarkall as from last comment of @esc it is clear to me that numba doesn't support outside package like langdetect,i was using langdetect in try block so it was not working,i guess the program i tried to execute on gpu can't be done using numba,thanks for your comment

[segatrade]

Hello!

I run:

try:
    xx_inv = np.linalg.inv(xx)
except np.linalg.LinAlgError:
    return [np.nan], [[np.nan, np.nan]]

It gives:

numba.core.errors.UnsupportedError: Failed in nopython mode pipeline (step: analyzing bytecode)
'try' block not supported until python3.7 or later

I have python3.6, but in my case this is requirement. Is any way to use try on py3.6 with numba or replacement for try?

Thank you!