In T236614#5858056, @Mpaa wrote:

and also wasted me a lot of time in the process of realizing that I need to implement this and re-download data.

Obviously developers and maintainers are not happy when bugs slip-through and users are annoyed. Anyhow, it does not seem right to me that you are complaining of having "wasted" your time, when people are donating their free time to carry on this project.

I just want to give my perspective on an issue in order for it to be better triaged:

1. pywikibot claims to support both Windows and Linux systems, so functionality must have a common denominator of "what is acceptable on all systems". If pywikibot doesn't support Windows anymore, probably we will need to remove the installation instructions for Windows to avoid confusion.
2. if we don't support Windows anymore, it's still beneficial to restrict filenames to the POSIX portable filename character set, which will guarantee that filename is portable between all Posix-like systems. For example, Kaggle, which is arguably one of the most popular platforms for data storage & processing, relies on the above-mentioned restrictions. Just a few days ago it was explained to me, that Kaggle does not support filenames not following POSIX portable filename character set. And probably, there are a lot of other use cases when this becomes unportable.

• Kaggle Notebook for processing dataset - https://www.kaggle.com/jacksoncrow/dataset-preprocessing
• Kaggle Notebook for training W2VV model - https://www.kaggle.com/jacksoncrow/w2vvtraining
• github repo - https://github.com/OlehOnyshchak/WikiImageRecommendation

Full dataset - https://www.kaggle.com/jacksoncrow/extended-wikipedia-multimodal-dataset
Full dataset with raw images - https://drive.google.com/file/d/1l0Oyv2Y6LmPGN3lP9MB6i8WWCinqkYPk/view?usp=sharing

Filed a bug about missing functionality of pywikibot while collecting additional data T238992. Currently, I applied an inefficient workaround.

Full dataset - https://drive.google.com/open?id=18i0D-N1J18UC1ebT9qbHZegKJQiKba5z
Will post a link with full dataset on Kagle.

FIled another defect, while finetuning image download T236614

@Dvorapa @Xqt, thanks for support and information!

@Xqt, thank you for the resolution! While pypi version still has this bug, I will switch directly to the master branch.

Created a truncated dataset for 500 articles(1.3 Gb) - https://www.kaggle.com/jacksoncrow/wiki-articles-multimodal
Full dataset for 5638 articles(14.2 Gb) is still uploading, will follow up with a link

I see, thanks. To make things more simple, I just reproduced the bug in a clean online environment with the latest pywikibot - https://jupyter.org/try
In other words, it allows you to reproduce it easily as well. You can see installation/demonstration of the bug in the attached file:

@Dvorapa, thank you for such a fast response!

@Aklapper, thank you for the clarification! Will report it properly

Sent email with found pywikibot bug for some articles (crashes and can't download some images)