Re: Contribution queries

[Oswald B. <os...@us...>]

On Wed, Apr 08, 2026 at 06:54:19PM +1000, Seth McDonald wrote:
>(0) Does the codebase use as reference any particular set of standards
>or implementations when choosing what libraries and functions to use?
>For example, does it only use standard system calls that are included in
>POSIX.1 (and if so, which version?), or that are available on Linux,
>FreeBSD, macOS, etc.?
>
i'm not overly conservative about function use. if it's been widely 
deployed on the major platforms for a while, it's good to go.

adding new library dependencies is a different story. the added value 
would have to be rather substantial.

>(1) When sending patches, do you prefer each single change as its own
>patch in an overall patchset, or multiple related changes squashed into
>a single patch?  I personally default to the former, but can do the
>latter if desired.
>
i think we are on the same page in principle.

>(2) Would you be open to accepting patchsets whose sole purpose is to
>refactor the code, such as for better modularity, extensibility,
>flexibility, or just readability?
>
if the improvement is substantial, i'm open to it. but i don't like 
pointless churn that just messes with git-blame. best you ask in 
advance.

>(3) Same as above, but for patches which mainly add function comments.
>I do actually have a specific reason for this one.  Many functions seem
>to have a loosely defined purpose and behaviour, such that calls to them
>sometimes depend on details of the function's implementation which need
>not stay the same over time.  Adding some comments outlining exactly
>what behaviour should or should not be relied upon would (I think) be
>beneficial for preventing such subtle misuse of functions.
>
i don't like verbosity, and it's in fact the reason why i haven't 
applied your last series yet ... i need to think of ways to condense it 
without losing the info i find valuable in principle.

comments have a very strong tendency to become stale.

for highly specialized functions with a single caller, having proper 
apidocs is arguably more of a liability than an asset. such call chains 
should be understood as a unit anyway, and should be modifiable with 
minimal overhead.

for more general utility functions, something doxygen-like makes sense 
in not entirely self-explanatory cases. i'm a fan of documenting 
constraints with assert()s, some of which you'll find in my more recent 
code. descriptions of object ownership are useful in non-obvious cases.

>(4) Would you prefer patchsets be sent in only after any previous
>patchset was reviewed and (if accepted) pushed upstream, or patchsets be
>sent in whenever they're ready?
>
i don't care. you need to do the rebasing ...