[mypyc] feat: constant folding for DictExpr [1/1]

[BobTheBuidler]

It should be possible to "constant fold" a LOC such as:

requests.get(some_url, headers={"k0": "v0", "k1": "v1"})

We can't simply create a dictionary constant since dicts are mutable and we don't know what's going to happen in the called function, but we can hold a template dict in memory and PyDict_Copy it when it's time for use. This skips the hashing and any resizing during construction, and simply takes the hash table from the existing dict instead, which will be (possibly much) quicker than constructing from scratch each time.

This PR does that.

[BobTheBuidler]

is there a reason we can't represent the type precisely? I think with a HashableLiteralValue defined before LiteralValue, we can do it pretty cleanly

[JukkaL]

I verified that this makes a simple micro-benchmark about 20% faster. However, this is quite a big and complex PR, and I'm not sure if the use case is common enough to support the "complexity budget". Can you show some real-world use cases where this would help?

[BobTheBuidler]

Nothing specific I can readily share but based on my own use case I see it being useful in any code that's making requests and sending consistent or quasi-consistent headers along with them, or code that creates a dict by starting from some base dict expr and modifying it with conditional logic below For me this is most meaningful for a high thruput internal API

…

On Tue, Oct 14, 2025 at 8:29 AM Jukka Lehtosalo ***@***.***> wrote: ***@***.**** commented on this pull request. I verified that this makes a simple micro-benchmark about 20% faster. However, this is quite a big and complex PR, and I'm not sure if the use case is common enough to support the "complexity budget". Can you show some real-world use cases where this would help? — Reply to this email directly, view it on GitHub <#19864 (review)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AQ3HIHR7CQ7Q2LVHXJQD44L3XTUCBAVCNFSM6AAAAACGXC4PYGVHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMZTGMZVGM2TGMRVG4> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[BobTheBuidler]

Also helps for functions that need their own little lookup tables that aren't worthy of being a module level constant. I have a bunch of these little lookup tables in [one repo](https://github.com/BobTheBuidler/yearn-treasury/) I maintain example: ```python def check_owner(x: LicensePlate) -> bool: owner = {"honda": "tom", "ford": "tim", "vw": "tam"}[x.car] return is_good(owner) ``` edit On Tue, Oct 14, 2025 at 8:33 AM Bob Buidler ***@***.***> wrote:

…

Nothing specific I can readily share but based on my own use case I see it being useful in any code that's making requests and sending consistent or quasi-consistent headers along with them, or code that creates a dict by starting from some base dict expr and modifying it with conditional logic below On Tue, Oct 14, 2025 at 8:29 AM Jukka Lehtosalo ***@***.***> wrote: > ***@***.**** commented on this pull request. > > I verified that this makes a simple micro-benchmark about 20% faster. > However, this is quite a big and complex PR, and I'm not sure if the use > case is common enough to support the "complexity budget". Can you show some > real-world use cases where this would help? > > — > Reply to this email directly, view it on GitHub > <#19864 (review)>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/AQ3HIHR7CQ7Q2LVHXJQD44L3XTUCBAVCNFSM6AAAAACGXC4PYGVHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMZTGMZVGM2TGMRVG4> > . > You are receiving this because you authored the thread.Message ID: > ***@***.***> >