PR#3703: RawHeader improvements -

koji

#3703 RawHeader improvements

Merged Mar 23, 2023 by tkopecek. Opened Feb 20, 2023 by mikem.

mikem/koji sig-header-fields into master

mikem commented Feb 20, 2023

While looking at #3690 I ended up using RawHeader.dump() to do some analysis, and this prompted me to fix a few minor issues there.

display signature header tags correctly
handle int alignment padding
avoid masking built-in next function
somewhat more informative output
remove strict rpmlib dependency

This also includes fixes from #3714

decode option
support addition datatypes in get: char, string array, i18n array
fix handling for integer types

Fixes https://pagure.io/koji/issue/3713

7 new commits added

flake8 fix
use surrogateescape to decode in RawHeader
decode option in RawHeader, handle integer counts
flake8 fix
attempt to decode strings in dump
also support get for unicode string arrays
RawHeader.get can return also string lists

Mar 13, 2023

mikem commented Mar 13, 2023

Note: this conflicts with #3690 because RawHeader.dump() will now return a list of ints for RPMTAG_LONGSIGSIZE.

1 new commit added

single opt for get, plus simple unit test

Mar 13, 2023

mikem commented Mar 13, 2023

So, I'm a little puzzled about what to do for scalar header values. The RawHeader class is deliberately a lower level tool for parsing rpm header structure, whereas rpm itself applies several smarter layers of interpretation. In particular, rpm treats some header fields as scalars and others as arrays without any indication in the header apart from the header tag. E.g. buildtime is a scalar int32 and changelogtime is an int32 array, both both are represented as RPM_INT32_TYPE in the actual header.

We could have RawHeader vary its behavior here based on key, but that would require embedding a bunch of tag data from rpmlib. It's doable, but we'd need to keep it up to date (and it doesn't look like this aspect of the data is exposed in the python api).

To further complicate things, the RawHeader code before this PR had a bug for int values. It only returned the first int one, effectively treating even the int arrays as scalars.

The compromise solution I have here is to add a single option to RawHeader.get as a convenience for callers that want to fetch a value they expect to be singular.

mikem commented Mar 13, 2023

@tkopecek @puiterwijk any thoughts on the above?

mikem commented Mar 13, 2023

It's also worth noting that while the rpm code treats RPM_I18NSTRING_TYPE as an array type like RPM_STRING_ARRAY_TYPE, all actual i18n tags are scalar (summary, description, group).

tkopecek commented Mar 14, 2023