fix(file source): fix regression in fingerprint calculations #8187
Conversation
0.14.0 included an upgrade to the `crc` crate, with associated updates to use the new API, however it appears to calculate different checksums. I put a comment here asking about how to calculate an equivalent checksum using the new API: mrhooray/crc-rs#62 (comment) This also adds an alias to ensure that checkpoints written by 0.14.0 will be able to be read by 0.15.0 when it is released. Fixes: #8182 Signed-off-by: Jesse Szwedko <[email protected]>
There was a problem hiding this comment.
The only issue I see with this is that it's going to be incompatible with checksums from 0.14.0. I'm not sure how best to resolve that, since calculating and comparing both seems like it'd be a big pain.
I would also add a simple test that compares the full serialized JSON of the checkpoints file to a fixed string, so that we'll get a failure and see a diff anytime it changes. Something like that should catch both the crc and the variant rename issues.
Blah, yeah, that's true, it will be incompatible with the changes in 0.14.0 now. I don't think there is a way to differentiate the checksums written by 0.13.1 vs. 0.14.0. We could try to calculate both checksums but it would be a pain as you note. Maybe we should just accept the regression, document it on the release page for 0.14.0, and include the tests to guard against it in the future. |
|
One potential option would be to do something similar to the existing |
|
Replaced by #8225 |
0.14.0 included an upgrade to the
crccrate, with associated updatesto use the new API, however it appears to calculate different checksums.
I put a comment here asking about how to calculate an equivalent
checksum using the new API:
mrhooray/crc-rs#62 (comment)
This also adds an alias to ensure that checkpoints written by 0.14.0
will be able to be read by 0.15.0 when it is released.
Fixes: #8182
Signed-off-by: Jesse Szwedko [email protected]