Natanael Copa authored 2 months ago and Timo Teräs committed 2 months ago

So far we only implement:

- version_validate
- version_compare
- version_match

fixes: #11062

The implementation had subtle error and allowing this is subject to
subtle errors in users due to the assumption of zero being there.

Just remove this, and the only place relying it needs to now do
a copy for now.

This can be used to create a subset of existing index.

E.g "abuild listpkg" could be used to create list of packages
of current repository and this list can be used to filter
existing index. Essentially removing packages from index
that have wrong version or which are not existent in repository
source.

Combined with --pkgname-spec with the original backing repository
URL, this can create an index with specific set of packages that
refer to upstream repository.

fixes #10873 for my use cases

avoids an apk_url_sanitize call and reduces memory usage

remove the implicit zero termination assumption

- take a maximum string length to allow usage with blobs
- use "test:" instead of "file://localhost/" for the tests
  which use local file path want it treated as network repository

Do not use repos[] for the cache, instead use a dedicated
struct apk_repository cache_repository for it.

Since OpenSSL 3.0, the functions EVP_sha256, EVP_sha512 etc. use
internal locking to accommodate internal changes to EVP_MD. Despite
being negligent, it still adds up when called repeatedly.

Switch to "explicit fetching" [1] to remove the unnecessary locking.
OpenSSL <3.0 and LibreSSL is unaffected by this.

[1]: https://docs.openssl.org/3.1/man7/crypto/#explicit-fetching

use instead bool/true/false or enum types

amend information about apkv3 details and align with current code

- pregenerate the needed repository urls
- get rid of apk_url_print and simplify url printing

Recommend --prune/--available if non-repository packages are seen.

fixes #10695
fixes #11050

return native error codes and map them in more detail in apk

fixes #10668

fixes commit 4c776eb7 db: make index refresh notices a progress note

Lot of complexity for very little value.

Make sure the pure virtual names (no auto selectable packages)
are resolved last. This allows dependency processing to uncover
additional constraints that might enable selecting them.

fixes #11059

also adjust test suite to test for package removal

fixes 6007db9c print: refactor progress printing API

Test case from jane400@postmarketos.org

fixes #11060

The 'fetch ...' lines of update are not really applet output, but
status notifications to indicate that something network related
is happening.

Since that cannot go to stdout as some applets can output files
directly to stdout it was made to go to stderr. That also is
slightly problematic as it can be interpreted as an error, and
if both stdout/stderr are redirected to different files/pipes
their relative order might get jumbled.

Convert these notices to a new progress note that follows the
progress output scheme: erase the output once it is no longer
relevant. And follow the --(no-)progress state so these get
automatically disabled if output is piped.

and hide it from the headers

nina authored 3 months ago and Timo Teräs committed 2 months ago

This lets VERSION contain additional info and without making
the library filenames messed up.

fixes #10808

Instead of deferring rendering progress bar to next progress update,
render it immediately after writing an output message.

- print progress based on istream bytes read:
  * usually network is slower than disk write speed
  * make consistent on the progress usage on all scenarios
- simpify things as progress is not passed to fs ops

- introduce _start and _end calls
- implicitly print the 100% on end call
- include "stage" in the _start call and relay it to progress-fd