[meeting] [joint] 07.06.2024

[deemp]

Danila's notes

Realistic example

• An example that doesn't evaluate to a constant.

Problem

• We don't have a realistic example.

Solution 1: try the multiplication table example

link

Problems:

• There are problems on the EO side (eoc unphi produces untestable programs eo#3213 (comment))
• We can only normalize.
• We can't normalize inside atoms (stdout).
• Hence, we can't reduce metrics.
• Hence, this example is bad.

Solution 2: artificial example

• Compose an EO program that can be normalized via user-defined rules #404

Solution 3: do more

• Evaluate arguments inside atoms. @yegor256 says it's uncool because normalizer isn't an evaluator.
• Rules for phi-expressions (YAML-based). @fizruk, I missed the details of the idea.

Tests

• Test partially normalized programs #405

Global object

• Reduce the object size. @fizruk

Anatoliy's notes

• Ivan gave feedback: EO examples are too simple, hence they fold into simple values. Need a bigger example.
  • Nikolay suggests to make up a larger example.
• According to Nikolay and Maxim, there is no syntatic support for (?)some representation of bytes(?).
  • Normalizer team had to roll back some stdout results.
• In general, we do not know if we can safely dataize some object. Counter example with multiplication table where normalization cannot optimize anything.
  • We will check how much normalization can help with multiplication table.
• Yegor suggests to postpone large example to later phases and now continue the work on normalizer.
• It is interesting to see if partial normalization preserves program semantics (Test partially normalized programs #405).
  • Such things should be added to tests.
• Global variables for rho might be implemented this week (Implement global variables for rho #408).
• Right now there are issues with XMIR (Run normalizer on the spring-fat archive with the optimization as a custom rule #379 (comment)). According to Vladimir, Normalizer team cannot do anything from their side to help. Waiting for a solution from EO team side ().
• Yegor wonders how they can keep typing information in EO. They could be kept as separate objects, however they might get destroyed by the normalization process. However, Yegor wishes to keep some semantic information from Java Bytecode (Keep typing information #409).
• For patent, we would need one non-generic rule. Some rule that performs a non-trivial optimization to include it into patent proposal (Plan a patent #197).