Use MemorySourceAccessor in DummyStore so writes work

[Ericson2314]

Motivation

This makes the dummy store a lot more useful. In particular, it opens the door to running complex integration-ish store tests entirely in memory as part of the units tests so we have a reference implementation to hold against other store implementations.

Context

This was originally made for #10874.

Add read-only setting to dummy:// store for back compat/

Priorities and Process

Add 👍 to pull requests you find important.

The Nix maintainer team uses a GitHub project board to schedule and track reviews.

[Ericson2314]

Note that per #13757,

nix/src/libfetchers-tests/git.cc

Lines 182 to 185 in d7612f3

	// TODO: Use dummy:// store with MemorySourceAccessor.
	Path storeTmpDir = createTempDir();
	auto storeTmpDirAutoDelete = AutoDelete(storeTmpDir, true);
	ref<Store> store = openStore(storeTmpDir);

is what we'd like to replace with this.

[nixos-discourse]

This pull request has been mentioned on NixOS Discourse. There might be relevant details there:

https://discourse.nixos.org/t/2025-09-17-nix-team-meeting-minutes-248/69508/1

[fzakaria]

Sick!

I actually think memory as a store name would be better as an alternative store.....

Dummy being "dumb" was kind of nice -- it was almost like "DevNullStore"

[xokdvium]

it was almost like "DevNullStore"

It still is. You can't actually query store paths from it.

[fzakaria]

@xokdvium maybe I mis read the description but I thought you could now.. that's why it needs "read-only" for back-compat.

You can write to it but not read ? Yet?..... ;P

[xokdvium]

You can write to it but not read

Sort of. You can't query path infos now and some of the stuff breaks. This is enough for the purposes of unit tests though.

[Ericson2314]

Oh oops I meant for it to support querying path infos? Or at least I didn't intentionally make that not work.

If we want to separate memory store from dummy store, that's fine for me too.

[xokdvium]

for it to support querying path infos

Then queryPathInfosUncached needs to be implemented?

[Ericson2314]

Ah yeah I remember now, I wanted a map with the values being pairs of valid path infos and memory source accessors, but the interface still expects a single source accessor for the entire store.

[fzakaria]

@Ericson2314 i'm down to pair on that.
I'd be really up to:

• add a new Store type "memory://"
• Remove the write from DummyStore (keep at NullInput and Nulloutput)
• Add read/write support to memory
• bonus: if we make memory:// use a sqlite memory database for the store (BLOB), we could trivially have it emit sqlite databases which could be a cool way to ship and reproduce bug.

[Ericson2314]

OH oops @fzakaria, I already started this!

[fzakaria]

@Ericson2314 you even had the sqlite idea too?!
(We are sync'd ?)

[Ericson2314]

Nah not that part. Keeping it it all pure C++