On Fri, Feb 06, 2026 at 10:40:14AM +0100, Fiona Ebner wrote: > Am 30.01.26 um 9:23 AM schrieb Arthur Bied-Charreton: > > On Thu, Jan 29, 2026 at 06:38:50PM +0100, Thomas Lamprecht wrote: > >> Am 21.01.26 um 11:34 schrieb Arthur Bied-Charreton: > >>> Add explicit length checks to ID validation functions to provide clearer > >>> error messages in case of length mismatches > >> > >> A few of these checks might be also done by setting the "maxLength" and/or > >> the > >> "minLength" property in the matching "register_standard_option" schema > >> definition. > >> Could you check that? Would be IMO better to reuse the schema > >> capabillities, and > >> that way, this limits would be also visible in the api schema directly. > >> > >> > > Hey Thomas, thanks for the feedback, I did not think about that. > > Not Thomas, but taking the freedom to respond :) > > > After looking into this, it seems like we could get rid of most of the > > parse_*_id functions > > (and probably others as well), but this would require changing > > PVE::JSONSchema::check_format > > Do you mean check_prop() here? > Yes I do, got the names confused -_- Thanks!
> > to check in the following order: > > > > `length -> pattern -> custom format fn` > > > > as opposed to currently: > > > > `custom format fn -> pattern -> length` > > > > In the JSON Schema validation's current implementation, the length > > parameters are effectively ignored > > due to the pattern/format function being checked beforehand. Do you see any > > issue with changing the > > validation order? > > AFAICS, changing the order does not change the execution flow. For any > violation, add_error() is called and then the function returns. So for > inputs with multiple violations, changing the order of checks just > changes which error message is present. When messages about length are > more telling to the user (and I think that's true in most cases), it is > an improvement if they are preferred. > That is what I thought/observed as well. > Do you have a good reason for also changing the order between the format > fn and pattern? Are there even cases where both are used? If there are > such cases, I guess it depends on the format fn implementation whether > the error message from there is more telling than "value does not match > the regex pattern". What do you think? I do not, I wrongly operated under the assumption that format fns would always have the kind of error message we are trying to get rid of. It is true that custom error messages are unlikely to be *less* telling than regex errors, the correct order is more length -> fn -> regex. Thanks for catching that! Now since a lot of the registered format functions only check values against their expected lengths and patterns, they could be fully replaced by registering schemas with register_standard_option instead of custom code. I will submit a v2 with these changes after I made sure that removing the registered formats does not result in any regressions. Of course if anyone already knows about some off the top of their head, I would be thankful for any tips :).
