[jira] [Comment Edited] (ARROW-16872) open_csv throws ArrowInvalid if csv does not end with a new line and is above 16384 lines

["Frederik Fabritius (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/ARROW-16872?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17557267#comment-17557267 ] 

Frederik Fabritius edited comment on ARROW-16872 at 6/22/22 6:37 AM:
---------------------------------------------------------------------

hm, it happened both in production on a google cloud compute vm debian-instance, and made that minimal reproduction test locally on my x86 mac.

When trying to run this in my dev environment, it worked this morning? But i found that it doesn't work with a fresh environment installed via `mamba` (a faster conda).

 

Here are steps to fully reproduce it:

```

mamba create --name test pyarrow=8 python=3.9 

conda activate test

python

```

 

And then in the python console, use the same reproduction test as before:

```

import pyarrow as pa

import pyarrow.csv

from io import BytesIO

for _ in pa.csv.open_csv(BytesIO('\n'.join(['review_id,filter_outcome'] + ['62593aaec7628b203bad4c6e,fabrication']*16385).encode())): pass

```

So it seems to only happen in a fresh environment, where the pa.csv.open_csv has not run successfully before. Could be a mamba issue, a conda issue, a packaging issue or a compilation/abi issue. Have not dug deeper into what is going on.

 


was (Author: JIRAUSER291373):
hm, it happened both in production on a google cloud compute vm debian-instance, and made that minimal reproduction test locally on my x86 mac.

When trying to run this in my dev environment, it worked this morning? But i found that it doesn't work with a fresh environment installed via `mamba` (a faster conda).

 

Here are steps to fully reproduce it:

```

mamba create --name test pyarrow=8 python=3.9 

conda activate test

python

import pyarrow as pa

import pyarrow.csv

from io import BytesIO

for _ in pa.csv.open_csv(BytesIO('\n'.join(['review_id,filter_outcome'] + ['62593aaec7628b203bad4c6e,fabrication']*16385).encode())): pass

```

So it seems to only happen in a fresh environment, where the pa.csv.open_csv has not run successfully before. Could be a mamba issue, a conda issue, a packaging issue or a compilation/abi issue. Have not dug deeper into what is going on.

 

> open_csv throws ArrowInvalid if csv does not end with a new line and is above 16384 lines
> -----------------------------------------------------------------------------------------
>
>                 Key: ARROW-16872
>                 URL: https://issues.apache.org/jira/browse/ARROW-16872
>             Project: Apache Arrow
>          Issue Type: Bug
>          Components: Python
>    Affects Versions: 7.0.0, 8.0.0
>            Reporter: Frederik Fabritius
>            Priority: Major
>              Labels: csvparser, open_csv
>
> `pyarrow.csv.open_csv` throws ArrowInvalid if csv does not end with a new line and is above 16384 lines. Tested with both pyarrow 7.0.0 and 8.0.0. Error seen both in production app and on developer laptop.
>  
> Here's a minimal case for reproducing the issue:
> ```python
> import pyarrow as pa
> import pyarrow.csv
> from io import BytesIO
> for _ in pa.csv.open_csv(BytesIO('\n'.join(['review_id,filter_outcome'] + ['62593aaec7628b203bad4c6e,fabrication']*16385).encode())): pass
> ```
>  
> Error is thrown: 
> ArrowInvalid: CSV parse error: Expected 2 columns, got 1: 



--
This message was sent by Atlassian Jira
(v8.20.7#820007)