Found bug in migration checksum

[bryanmcgrane]

I noticed there is an issue with the checksum generation code found here:

prisma-engines/migration-engine/connectors/migration-connector/src/lib.rs

Line 88 in c1113c6

/// Compute the checksum for a migration script, and return it formatted to be human-readable.

I noticed the issue when inspecting the _prismia_migration tables that are generated in my database.

It appears there is an issue with converting from the bytes to a hex string, where 0's are sometimes being stripped out of the final checksum string. It's relatively simple to replicate the issue. Below is an example (apologies in advance, I'm not a Rust developer):

/// Prisma checksum
checksum("hello")

/// Result is 2cf24dba5fb0a3e26e83b2ac5b9e29e1b161e5c1fa7425e7343362938b9824
/// Note the length is only 62 characters

#!/bin/bash
echo -n hello | shasum -a 256

# Result is 2cf24dba5fb0a30e26e83b2ac5b9e29e1b161e5c1fa7425e73043362938b9824
# Note the length is the correct 64 characters

Characters that are missing from Prisma's checksum
2cf24dba5fb0a30e26e83b2ac5b9e29e1b161e5c1fa7425e73043362938b9824

Edit:
I dug a little deeper and I found the issue was with this line where bytes are converted to hex strings.

prisma-engines/migration-engine/connectors/migration-connector/src/lib.rs

Line 110 in c1113c6

write!(checksum_string, "{:x}", byte).unwrap();

write!(checksum_string, "{:x}", byte).unwrap(); needs to have a 0 padding built in for bytes with leading 0's.

[tomhoule]

Hi @bryanmcgrane! Good catch, thanks for reporting this, I'll look into it asap. We'll have to be careful to still accept the bad checksums for now, unfortunately (while fixing the new ones), or it will break existing users.

[tomhoule]

PR is here: #1889

I think we should merge after next release (Tuesday), since current checksums aren't completely broken, and that will give us some time to test before releasing that change to users.