Give the relay a shared store, an owner, and a lifetime longer than one process by thisisjun786 · Pull Request #8 · thisisjun786/codex-relay-workflow

Write down who owns the relay's state and its daemon

thisisjun786

committed

62ec00a

Let every participant prove which store it is actually using

thisisjun786

committed

1e64a12

Mark the state-selection and store-identity contract as shipped

thisisjun786

committed

035834c

Give the relay an owner, and make it refuse to guess who that is

thisisjun786

committed

98cc235

Carry an assignment past the lifetime of the process serving it

thisisjun786

committed

59197a8

Treat two names for one socket as one operating scope

thisisjun786

committed

d48c8e4

Answer the ownership findings review raised on this branch

thisisjun786

committed

1610302

Answer the second review round on ownership and identity

thisisjun786

committed

7527384

Require a live pid before start calls a launch successful

thisisjun786

committed

5d0924f

Answer the third review round on ownership, side effects and readiness

thisisjun786

committed

37118cd

Answer the fourth review round on abandoned children and unreadable answers

thisisjun786

committed

ca4f3f6

Answer the fifth review round on shared intent, stale worker reads and a fresh store

thisisjun786

committed

f522993

Answer the sixth review round on the record the worker must not write

thisisjun786

committed

c6fc0d1

Answer the seventh review round on shared intent and an absent boot

thisisjun786

committed

0752a69

Answer the eighth review round on enable, a missing boot and a blocked socket

thisisjun786

committed

e389908

Answer the ninth review round on an upgrade that would have lost a store

thisisjun786

committed

f155c88

Answer the tenth review round on a directory that is not a store

thisisjun786

committed

c138b80

Merge remote-tracking branch 'origin/dev' into codex/jun-103-shared-relay

thisisjun786

committed

639bde8

Close three ownership races the lock could already have settled

thisisjun786

committed

a3f3dc0

Four more places where a second launch was not accounted for

thisisjun786

committed

9462291

Two launches with no launch id are still two launches

thisisjun786

committed

903a2ce

Order launch identity by stability, so a normal exit is not a replacement

thisisjun786

committed

d9d688c

Two refinements to identity and discovery, both from review

thisisjun786

committed

471e50e

Adopting neither store was not the visible failure I argued it was

thisisjun786

committed

5ee8b6d

Four more ways a lifecycle command acted on something it had not identified

thisisjun786

committed

f416667

Three more from round six, two of them about who was holding what

thisisjun786

committed

2f36f02

Round seven: a stop that could erase a live launch, and two smaller things

thisisjun786

committed

b64ceff

Two windows my last round left open

thisisjun786

committed

2c516bc

One rule for the stop record, after three narrower ones each left a window

thisisjun786

committed

d243e63

open() reports EACCES too, and that is not contention

thisisjun786

committed

5681e26

Navigation Menu

Give the relay a shared store, an owner, and a lifetime longer than one process - #8

Give the relay a shared store, an owner, and a lifetime longer than one process#8
thisisjun786 merged 30 commits into
devfrom
codex/jun-103-shared-relay

Commits on Sep 16, 2026

Write down who owns the relay's state and its daemon

Let every participant prove which store it is actually using

Mark the state-selection and store-identity contract as shipped

Give the relay an owner, and make it refuse to guess who that is

Carry an assignment past the lifetime of the process serving it

Treat two names for one socket as one operating scope

Answer the ownership findings review raised on this branch

Answer the second review round on ownership and identity

Require a live pid before start calls a launch successful

Answer the third review round on ownership, side effects and readiness

Answer the fourth review round on abandoned children and unreadable answers

Answer the fifth review round on shared intent, stale worker reads and a fresh store

Answer the sixth review round on the record the worker must not write

Answer the seventh review round on shared intent and an absent boot

Answer the eighth review round on enable, a missing boot and a blocked socket

Answer the ninth review round on an upgrade that would have lost a store

Answer the tenth review round on a directory that is not a store

Merge remote-tracking branch 'origin/dev' into codex/jun-103-shared-relay

Close three ownership races the lock could already have settled

Four more places where a second launch was not accounted for

Two launches with no launch id are still two launches

Order launch identity by stability, so a normal exit is not a replacement

Two refinements to identity and discovery, both from review

Adopting neither store was not the visible failure I argued it was

Four more ways a lifecycle command acted on something it had not identified

Three more from round six, two of them about who was holding what

Round seven: a stop that could erase a live launch, and two smaller things

Two windows my last round left open

One rule for the stop record, after three narrower ones each left a window

open() reports EACCES too, and that is not contention