Simon Conseil

Indeed, even if you ignore some keyword, `diff.report()` still reports that the number of keywords is different (because `diff_keyword_count`) ``` In [5]: print(diff.report()) Headers have different number of cards: a:...

I understand the idea but I don't know if the current behavior was on purpose. Maybe @embray can enlighten us ? :pray:

> The fill_value attribute is basically just saying what to use when making a filled version of the table via the filled() method. Thus it should not (AFAIK) be directly...

> It does not seem to round trip for float or string columns. I spent a fair bit of time on #7481 to provide and optional work-around for these problems,...

@SaOgaz - probably not the same issue, but it's hard to say without a minimal example to reproduce ;). I think it deserves its own issue.

#7481 added a new option to serialize the mask without relying on `fill_value`/`TNULL` (`t.write(... serialize_method='data_mask')`), however using the example above and not using this option it seems that now `fill_value`...

I agree the current proposed change is not desirable, `fitsheader` should not modify the keyword values. The non-`fitsort` mode (`--table`) already accepts a format parameter, set to `ascii.fixed_width` by default....

@mromanie - you should push your changes to your branch (https://github.com/mromanie/astropy/tree/patch-1), or use the github editor on that branch. But I see that there various points that could be improved...

This is to return the dev version when running astropy from a git checkout: #10774

For offline usage something like [dash](https://kapeli.com/dash) or [zeal](https://zealdocs.org/) works well (it easy to do, running `doc2dash` on Sphinx's output, and submit to https://github.com/Kapeli/Dash-User-Contributions). PDF is useless imho.