epage
commented
5 years ago Good idea. While "golden state" tests are brittle, they can be useful and making it easier to update them would be a big help.
We'll also want to look into this for #33. The main impact of #33 on this issue is ensuring we have good names for both of them.
I'm assuming the snapshots would be kept under revision control.
Options for where to store
- Fixed location under
tests/fixtures. - Require the path be specified
- Use a builder for this (e.g.
predicate::eq_snapshot!().from("tests/fixtures")) with a default location
Next, what is the snapshot's filename
- Manually specify it
- Use the file and line macros
- This won't work if the macro is used in a test helper
- This is brittle because the file names will change with the smallest test change
- If all snapshots are dumped in an isolated directory, the workflow can just be to
git rm * && cargo test && git add *
- Using a function name macro when / if its available (see https://github.com/rust-lang/rfcs/pull/466 / https://github.com/rust-lang/rfcs/issues/1743) would be less brittle but only allow one per test
- Use a builder for specifying it with a default
eq_snapshot would probably want a .utf8() like eq_file so we know when can provide more useful failure messages.
Any other API considerations?
Did we ever talk about snapshot tests? I don't recall, and I didn't find an issue in this or the assert_{cmd,cli} repos.
The idea is this: Instead of manually writing
predicate::path::eq_file(Path::new("some.file")).utf8().unwrap();, you'd have apredicate::eq_snapshot!()macro that expands to a uniq filename predicate with the ability to create that file with the content it gets on the first test run or when usingRECORD_SNAPSHOTS=1or something similar.