fix[codegen]: fix some hardcoded references to STORAGE location

[cyberthirst]

What I did

• fixes several hardcoded references to STORAGE where we actually also care about TRANSIENT
• also fixes existing issues:
  • Optimizations Missing for Transient Storage #4017
  • Missing Constancy Check for Transient Storage #4018
  • Missing check for TRANSIENT in codegen #4020

How to verify it

• check coverage that the new updates are tripped

Commit message

fix some hardcoded references to `STORAGE` when `TRANSIENT` is also
possible. these are not breaking changes semantically, but should enable
the correct optimization codepath for transient storage. this commit
also results in better codepaths taken for the IMMUTABLES location
(`iload`/`istore`), resulting in smaller bytecode sometimes for
constructors.

Description for the changelog

Cute Animal Picture

EDIT: added references to the issues

[cyberthirst]

these i didn't update yet, but atleast the part for key is sus

• vyper/vyper/semantics/types/subscriptable.py

  Line 87 in 9428196

  key_type = type_from_annotation(k_ast, DataLocation.STORAGE)

• vyper/vyper/semantics/types/subscriptable.py

  Line 91 in 9428196

  value_type = type_from_annotation(v_ast, DataLocation.STORAGE)

EDIT:
also, maps are now instantiable in transient, so we should atleast update that

[charles-cooper]

why isn't the codecov bot running?

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 93.54839% with 2 lines in your changes are missing coverage. Please review.

Project coverage is 88.51%. Comparing base (0ba1c62) to head (daa8444).

❗ Current head daa8444 differs from pull request most recent head da66218

Please upload reports for the commit da66218 to get more accurate results.

Files	Patch %	Lines
vyper/codegen/core.py	83.33%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #4015      +/-   ##
==========================================
- Coverage   91.04%   88.51%   -2.53%     
==========================================
  Files         106      106              
  Lines       15309    15302       -7     
  Branches     3369     3369              
==========================================
- Hits        13938    13545     -393     
- Misses        933     1247     +314     
- Partials      438      510      +72     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[charles-cooper]

i tested several large contracts i have (CurveCryptoMathOptimized3-0.4.0.vy, VaultV3-0.4.0.vy, CurveStableSwapNG-0.4.0.vy) as well as everything in the examples/ directory and compared the sha256sum of runtime bytecode to master, and found no change.

there were some changes (improvements) to initcode, because after this commit, some more optimized codepaths for iload/istore are taken.

[cyberthirst]

feel hesitant to merge before we properly review coverage.

like do we hit all combinations in all paths where copy_opcode_available is used (esp. here return left.location == MEMORY and right.location.has_copy_opcode)?

[cyberthirst]

i'd prefer positional here

[charles-cooper]

feel hesitant to merge before we properly review coverage.

like do we hit all combinations in all paths where copy_opcode_available is used (esp. here return left.location == MEMORY and right.location.has_copy_opcode)?

i think we do. but codecov is not really aggregating the coverage reports properly. i think we should just double check the logic and not worry about it, i think those branches are actually getting hit (just not getting reported by codecov)