After a discussion, I was really curious that our panics are supposed to be almost free - and while trusting that statement, it was really hard to believe - so I wanted to see for myself - knowing that an experiment is the most robust source of this kind of information - testing that in practice.
So I wrote a benchmark comparing various ways of reporting errors, also testing them both at 'shallow' and 'deep' stack traces (adding 200 additional frames) - to see how stack depth affects them, if at all.
The panics are indeed blazing fast! Kudos to the engine team. However, it seems that our dataflow errors are relatively slow (and we tend to use them _more_ than panics and want to be using them more and more). This uncovers a possible optimization opportunity. Can we make them as fast as panics??
Analysis of the benchmark results in comment below.
- Fixes#7354
- And also closes#7712
- Refactors how we handle numeric ops - ensuring that the 'kernels' are placed all in one place and selected based on storage types.
- Closes#6111
- Aligns semantics of handling Mixed columns.
- Now, if an operation like `iif` or `fill_nothing` is given a `Mixed` column, the result will also be `Mixed` regardless of the `inferred_precise_value_type`.
- Enables a few old tests that were pending but could be enabled since the types work is advanced enough.
- Update list of groups to agreed list.
- Lower case `ALIAS` names to be consistent with function names.
- Add `GROUP` to methods.
- All constructors and functions have doc comments.
- Correct a few typos (e.g. `PRVIATE`).
- Mark some more things as `PRIVATE`.
- Use `ToDo:` and `Note:` consistently.
- Order tags in doc comment.
# Important Notes
We don't have all the doc comments on types and will want to add them in future,
- Closes#5159
- Now data downloaded from the database can keep the type much closer to the original type (like string length limits or smaller integer types).
- Cast also exposes these types.
- The integers are still all stored as 64-bit Java `long`s, we just check their bounds. Changing underlying storage for memory efficiency may come in the future: #6109
- Fixes#7565
- Fixes#7529 by checking for arithmetic overflow in in-memory integer arithmetic operations that could overflow. Adds a documentation note saying that the behaviour for Database backends is unspecified and depends on particular database.
The added benchmark is a basis for a performance investigation.
We compare the performance of the same operation run in Java vs Enso to see what is the overhead and try to get the Enso operations closer to the pure-Java performance.
