Intern string representation of operators and some other symbolic literals

[abebus]

Feature or enhancement

Proposal:

Python currently interns certain strings, such as keywords and some ASCII/Unicode characters, as well as module-specific strings. I propose extending this interning mechanism to the string representations of operators (e.g., "+=", "==", "|=").

Rationale:
Interning these strings could improve performance, particularly in code parsing workflows, by:

• Reducing memory overhead for repeated operator strings.
• Accelerating string comparisons (e.g., during AST construction or bytecode generation).

Target Symbols:
The following multi-character syntactic literals (with len() > 1) are candidates for interning:

# Syntax literals  
'...', '->' 

# Operators  
'**', '//', '==', '!=', '>=', '<=', ':=',  
'+=', '-=', '*=', '/=', '//=', '%=', '**=',  
'<<', '>>', '<<=', '>>=', '&=', '|=', '^='  

# And maybe character sequence that used in REPL?
'>>>'

Proof of Concept:
A preliminary implementation is available here, demonstrating the feasibility of this change.

Considerations:

• The change would be low-risk, as it targets immutable, statically known strings.
• The impact on startup time and memory usage should be negligible, given the small set of operators.

Would this be a worthwhile optimization for CPython? I’d appreciate feedback on the idea and the PoC.

Has this already been discussed elsewhere?

This is a minor feature, which does not need previous discussion elsewhere

Links to previous discussion of this feature:

No response

[StanFromIreland]

Are you able to provide benchmarks?

[abebus]

Thanks for the feedback! I’m currently looking into how to properly benchmark this. I haven’t run the full pyperformance suite yet, but I’m planning to get familiar with it and set it up properly.

In the meantime, would some synthetic microbenchmarks (e.g. repeated AST parsing or string comparison tests) be acceptable as a starting point? I can begin with those while I work toward running more representative, full-suite benchmarks.