Hi Baptiste, On 07/08/23 at 22:07 +0200, Baptiste Beauplat wrote: > Hi Lucas, > > On 2023-08-03 10:30, Lucas Nussbaum wrote: > > duck-as-a-service (duck.debian.net) has been broken for a long time, > > and > > the corresponding UDD importer is broken as well (see #949009, > > #963887). > > In the meantime, duck continued evolving (was rewritten?) and is now > > checking a lot more places for URLs. > > > > It would probably be useful to re-create a way to provide duck > > results > > as a service, based on UDD, similarly to what is done for upstream or > > lintian data. > > > > Ideally, this would be done in cooperation with the duck maintainer > > to > > do the following changes: > > - in duck, separate the logic to get URLs from sources, from the > > logic > > to check those URLs (for example, allow dumping a list of URLs, and > > also using a list of URLs as source) > > - in duck, provide machine-readable outputs (JSON?) > > Currently duck has two features which can help us: > > - The `-n` switch, which gets all URLs and prints them to stdout > - The `-l filename` switch, which takes a file with one URL per line > and checks them > > Theoretically, what's missing in only a `--json` switch, which would > change the output from console/text to JSON. > > But, as I see it, the `-l` argument is limited in two aspects: > > - It provides only the URL, loosing the checker type which is used to > select what kind of validation will be performed. > > For instance, a https://salsa.debian.org/rfrancoise/tmux.git of type > VCS-Git would be tested as a standard URL in the `-l` context, instead > of a git repository. > > - It requires a file > > I'm thinking of implementing a new JSON specific input format > (`--input-json`?), including the two information, which would read from > stdout instead of a file. > > The format would be as simple as: > > ```json > [ > {"type": "VCS-Git", > "url": "https://salsa.debian.org/rfrancoise/tmux.git";, > "filename": "debian/control", # optional key > "line_number": 10}, # optional key > ... > ] > ``` > > Following this logic, the output format for checking URLs would be the > same, as to have `duck --json -n | duck --input-json` working. > > The JSON result would hold an additional dictionary for each URL > entries > named "result", described as follows: > > ```json > [ > {"type": "VCS-Git", > "url": "https://salsa.debian.org/rfrancoise/tmux.git";, > "filename": "debian/control", # optional key > "line_number": 10, # optional key > "result": { > "state": 0, # 0 for OK, 1 for Error, 2 for Information > "detail": "Informative message", > "certainty": "possible" # optional key > }}, > ... > ] > ``` > > Let me know what you think of it.
That would be perfect! In the context of UDD, I will probably implement that as two tables: - one to store the mapping between source packages and urls (source, version, url, type, filename, line_number) which would be updated when a new source version gets uploaded - one to store the status of urls (url, type, result, timestamp of last check) which would be updated with a retry policy to be defined I would not use (filename, line_number) in the input of the URL testing part. The reason for that design is that it will easily allow to gather the status for several versions of the package (testing + unstable + experimental for example), while not duplicating the checks for URLs. Lucas
