Skip to content

5.0.1: fixes from the review of 5.0.0 - #208

Merged
compot-dev merged 3 commits into
masterfrom
qodo-202-fixes
Nov 21, 2026
Merged

compot-dev merged 3 commits into
masterfrom
qodo-202-fixes

Conversation

@compot-dev

Copy link
Copy Markdown
Collaborator

Fixes from the Qodo review of #202 (5.0.0); version 5.0.1.

Finding Fix
2 Client isolation turned on by a reload ends the network's mesh sessions: a direct path set up before would carry packets past the server. They come back and get no peer keys.
3 A TAP client keeps a given MTU (--mtu, mtu=, the server's CONFIG) instead of the LAN's 1500: mtu_given in the packet open parameters.
6, 8 A reload never keeps a client in a network it did not join: a slot that takes another network, or whose mode changes (mode = tap, addresses = none), ends its old sessions, key and PAKE ones too (a generation per slot).
7 A reload takes the gateway alias off the tunnel also for addresses = none or another pool size.
10 camex.ko: an skb queued while the offloads were on is linearised before an ordinary read (CAMEX_IOC_VNET turned off).
11 known_servers: one writer at a time (<file>.lock, flock), through a mkstemp file instead of a fixed .new.
12 --bridge-mode nat: a remote device not heard for 10 minutes is not sent to any more, and its neighbour entry goes.

1 and 9 are #206 and #207; 4 and 5 are answered in #202.

Tests, locally in the CI's images: all 25 e2e tests of the workflow (test-e2e-tap.sh now also checks that --mtu 1420 stays on a TAP device), camex.ko in QEMU (ubuntu:24.04: VM TEST PASS, version 5.0.1), Windows x86_64 and i686 under Wine, ci/build-musl.sh arm with the unit tests under qemu-arm, make test.

From the Qodo review of #202:
- Reload: a network slot that takes another network, or whose mode
  changes (tap, addresses = none), ends its old sessions (a generation per
  slot, netw.c); a client of a removed network no longer stays on in the
  one that took its slot, with its key (2, 6, 8).
- Reload: client isolation turned on ends the network's mesh sessions, so
  no direct path set up before goes on past the server (2).
- Reload: the gateway alias comes off the tunnel also for addresses =
  none or another pool size (7).
- A TAP client keeps a given MTU (--mtu, mtu=, CONFIG) instead of 1500:
  mtu_given in the packet open parameters (3); test-e2e-tap checks it.
- known_servers: one writer at a time (<file>.lock, flock) through a
  mkstemp file, not a fixed .new (11).
- --bridge-mode nat: a remote device not heard for 10 minutes is not sent
  to any more, and its neighbour entry goes (12).
- camex.ko: an skb queued while the offloads were on is linearised before
  an ordinary read (10).
1 and 9 are issues #206 and #207.
@qodo-free-for-open-source-projects

qodo-free-for-open-source-projects Bot commented Nov 21, 2026 •

Copy link
Copy Markdown

PR Summary by Qodo

Fix reload session isolation and networking regressions in 5.0.1

🐞 Bug fix 🧪 Tests 📝 Documentation 🕐 40+ Minutes

Grey Divider

AI Description

• End stale network and mesh sessions on reload, and reconcile tunnel gateway aliases.
• Preserve explicit TAP MTUs; fix kernel reads, bridge NAT expiry, and known-server writes.
• Add TAP MTU regression coverage and document the 5.0.1 fixes.
Merge Risk 🟡 Medium

Blast radius 🔴 High 💥 5 capabilities

  • 5 affected capabilities span independent functional scopes: client_server_key_persistence, kernel_packet_transport, named_network_server, nat_bridge_forwarding.
  • The generation is assigned to the named-network slot in netw_sync and consumed by server_networks_apply when deciding whether the client's existing session remains valid.
  • Mesh isolation uses the same session lifecycle and network membership boundary, so these are related outcomes of the bounded named-network integration rather than separate technical capabilities.
  • The alias is the server-side gateway address for the same named-network pool managed by netw_sync and server_networks_apply.
  • Its outcome is distinct from client session admission, but it remains within the same bounded named-network integration.
  • The changed lock and atomic rewrite protect the single end-to-end behavior of retaining learned server keys for future client starts.
  • Concurrent writers are multiple triggers of that same bounded persistence integration.
  • remote_by_ip is the consumer used by nat_in to select the destination remote, while remote_expire maintains the corresponding neighbour state.
  • These are producer/consumer hops of one LAN-to-remote forwarding capability.
  • The changed code is on the single ordinary character-device read path that delivers queued packets to the userspace tunnel consumer.
  • It does not introduce a separate functional scope from the queue and copy operation.
Sources

src/camex-k.c[669-675]

    /* A packet queued while the offloads were on (CAMEX_IOC_VNET turned
     * off since) may be in pages: copy_to_user() of skb->data would read
     * past its linear part */
    if (skb_linearize(skb)) {
        dev_kfree_skb(skb);
        return -ENOMEM;
    }

src/netw.c[479-486]

        /* Another network in the slot, or one that was not active, or its
         * mode changed: its old sessions do not carry over (server.c) */
        if (next[s].gen == 0U || !netw_table[s].active ||
            memcmp(netw_table[s].id, conf[j].id, CAMEX_NET_ID_LEN) != 0 ||
            netw_table[s].tap != conf[j].tap ||
            netw_table[s].noaddr != conf[j].noaddr) {
            next[s].gen = netw_gen_next();
        }

src/server.c[5185-5190]

        /* The same network, in the same mode: not one that took its slot
         * on this reload (netw.c, gen) */
        if (netw_table[n].active && c->net_gen == netw_table[n].gen &&
            (c->ip_be == 0U || netw_has(n, c->ip_be)) &&
            (c->pending_ip_be == 0U || netw_has(n, c->pending_ip_be))) {
            continue;
Behavioral complexity 🟡 Medium 🧩 Interacting branches, Coordinated invariants

  • Before: sessions were retained using network slot and address checks, while alias removal considered only network disappearance or gateway movement.
  • After: slot generations, network mode/address flags, isolation state, client generation state, and alias prefix must agree across synchronization and application.
  • New correctness conditions: a generation must change exactly when a slot's identity or relevant mode changes; an old client generation must never match a new occupant; isolation must only end applicable mesh sessions on the off-to-on transition; and alias cleanup must precede pool/address replacement.
  • These are newly interacting reload states, not merely reuse of an existing helper.
Sources

src/netw.c[479-486]

        /* Another network in the slot, or one that was not active, or its
         * mode changed: its old sessions do not carry over (server.c) */
        if (next[s].gen == 0U || !netw_table[s].active ||
            memcmp(netw_table[s].id, conf[j].id, CAMEX_NET_ID_LEN) != 0 ||
            netw_table[s].tap != conf[j].tap ||
            netw_table[s].noaddr != conf[j].noaddr) {
            next[s].gen = netw_gen_next();
        }

src/server.c[5185-5190]

        /* The same network, in the same mode: not one that took its slot
         * on this reload (netw.c, gen) */
        if (netw_table[n].active && c->net_gen == netw_table[n].gen &&
            (c->ip_be == 0U || netw_has(n, c->ip_be)) &&
            (c->pending_ip_be == 0U || netw_has(n, c->pending_ip_be))) {
            continue;
Potential damage 🔴 High 🩹 Confidentiality

  • If a changed generation or isolation check incorrectly preserves an old session, a client could retain access through a direct mesh path or a slot now representing another network.
  • That is a plausible cross-network confidentiality boundary failure.
  • If the check is instead overly strict, the consequence is temporary session loss; the serious classification covers the more consequential reachable fault.
  • Reverting code stops future incorrect decisions, but an already-retained session must be explicitly ended or naturally reconnect.
  • Any packets or access exposed before termination cannot be recalled; recovery requires the affected session to be revalidated or disconnected.
Sources

src/netw.c[479-486]

        /* Another network in the slot, or one that was not active, or its
         * mode changed: its old sessions do not carry over (server.c) */
        if (next[s].gen == 0U || !netw_table[s].active ||
            memcmp(netw_table[s].id, conf[j].id, CAMEX_NET_ID_LEN) != 0 ||
            netw_table[s].tap != conf[j].tap ||
            netw_table[s].noaddr != conf[j].noaddr) {
            next[s].gen = netw_gen_next();
        }

src/server.c[5185-5190]

        /* The same network, in the same mode: not one that took its slot
         * on this reload (netw.c, gen) */
        if (netw_table[n].active && c->net_gen == netw_table[n].gen &&
            (c->ip_be == 0U || netw_has(n, c->ip_be)) &&
            (c->pending_ip_be == 0U || netw_has(n, c->pending_ip_be))) {
            continue;
Assessed commit e268572
Diagram

graph TD
  C["Network config"] --> S["Slot sync"] --> G{"Identity or mode changed?"}
  G -->|yes| N["Advance generation"] --> A["Apply reload"] --> D["Drop stale sessions"]
  G -->|no| A
  A --> I{"Isolation newly enabled?"} -->|mesh clients| D
  A --> T["Reconcile aliases"]
Loading
High-Level Assessment

The following are alternative approaches to this PR:

1. End every network session on reload
  • ➕ Simpler session-validity rule; avoids tracking slot generations.
  • ➖ Disconnects unaffected clients on routine reloads and forces unnecessary reauthentication.

Recommendation: Keep the targeted generation and isolation checks: they invalidate sessions that may no longer be authorized without disrupting unchanged networks. Review the known-servers lock failure path separately, since proceeding without an acquired lock would not guarantee serialized writes.

Files changed (14) +170 / -9

Bug fix (11) +139 / -8
bridge.cExpire stale bridge NAT destinations +24/-3

Expire stale bridge NAT destinations

• Stops resolving a remote device by IP after its age limit and removes its associated neighbour entry.

src/bridge.c

camex-k.cLinearize queued packets before ordinary reads +7/-0

Linearize queued packets before ordinary reads

• Linearizes an skb before copying its data to userspace when the ordinary read path handles a packet queued with offloads enabled.

src/camex-k.c

camex.cCarry explicit MTU status through tunnel setup +5/-0

Carry explicit MTU status through tunnel setup

• Propagates the MTU-given flag into packet-open parameters, retains it during client reconfiguration, and marks a supplied global MTU as explicit.

src/camex.c

client.cSerialize Unix known-server file rewrites +43/-2

Serialize Unix known-server file rewrites

• Uses a separate flock lock file and a unique mkstemp file for Unix read-modify-rename writes. The Windows path retains its existing fixed temporary filename.

src/client.c

config.cMark profile MTUs as explicitly supplied +1/-0

Mark profile MTUs as explicitly supplied

• Sets the MTU-given flag when a server profile supplies an MTU, allowing TAP setup to preserve that value.

src/config.c

netw.cVersion network slots across reloads +23/-0

Version network slots across reloads

• Assigns a new generation when a slot changes network identity or address mode, and when a network is created. Preserves the old alias prefix when a slot changes hands so the server can reconcile its gateway address.

src/netw.c

netw.hTrack network generation and applied state +5/-0

Track network generation and applied state

• Adds fields for the slot generation, applied isolation state, and installed gateway alias prefix.

src/netw.h

packet.cDefault only unspecified TAP MTUs to 1500 +1/-1

Default only unspecified TAP MTUs to 1500

• Changes TAP device setup to use the LAN MTU only when no MTU was explicitly supplied.

src/packet.c

packet.hExpose explicit MTU status to packet backends +2/-0

Expose explicit MTU status to packet backends

• Adds an MTU-given field to packet-open parameters so TAP setup can distinguish a default from a supplied value.

src/packet.h

server.cInvalidate sessions and aliases affected by reloads +27/-2

Invalidate sessions and aliases affected by reloads

• Records each client's network generation during initialization and restores it on failed initialization. Revalidation and reload handling reject stale generations, end mesh sessions when isolation becomes active, and remove gateway aliases when addressing or pool prefix changes.

src/server.c

server.hRecord the network generation joined by a client +1/-0

Record the network generation joined by a client

• Adds a generation field to server client entries for checking session membership after network reloads.

src/server.h

Tests (1) +7 / -0
test-e2e-tap.shAssert explicit TAP MTU survives client startup +7/-0

Assert explicit TAP MTU survives client startup

• Adds an end-to-end check that a TAP client started with --mtu 1420 has a 1420-byte device MTU.

docker/test-e2e-tap.sh

Documentation (1) +23 / -0
CHANGELOG.mdDocument the 5.0.1 fixes +23/-0

Document the 5.0.1 fixes

• Adds release notes for the reload, TAP MTU, known-server persistence, bridge NAT, and kernel read fixes.

CHANGELOG.md

Other (1) +1 / -1
version.hSet the release version to 5.0.1 +1/-1

Set the release version to 5.0.1

• Changes the compiled default version string from 5.0.0 to 5.0.1.

src/version.h

@qodo-free-for-open-source-projects

qodo-free-for-open-source-projects Bot commented Nov 21, 2026 •

Copy link
Copy Markdown

Code Review by Qodo

🐞 Bugs (5) 📘 Rule violations (0) 📎 Requirement gaps (0) 🎨 UX issues (0) 🔗 Cross-repo conflicts (0) 📜 Skill insights (0)

Grey Divider


Action required

1. Isolated clients keep direct peer paths 🐞 Bug ⛨ Security
Description
server_networks_apply() calls server_end_client() when isolation turns on, but ending the server
session neither revokes the clients’ mesh keys nor disables their direct peer connections. After the
reload, clients can keep sending directly without a valid server session, and reconnecting does not
stop mesh because the server still advertises it for that network.
Code

src/server.c[5182]

+            server_end_client(c);
Evidence
The new reload branch ends only the server entry. Direct sends precede the client’s server-session
validity check, while the server’s CONFIG construction can continue to advertise mesh after
isolation is enabled; the client’s live CONFIG comparison does not check mesh.

src/server.c[5174-5184]
src/server.c[823-851]
src/client.c[1059-1082]
src/server.c[1288-1322]
src/client.c[594-628]

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
Ending server sessions does not clear clients’ existing direct peer keys, so enabling isolation leaves direct traffic possible.
## Fix Focus Areas
- src/server.c[5174-5184]
- src/server.c[1288-1322]
- src/client.c[594-628]
- src/mesh.c[204-225]
## Recommended Fix
Make the isolation transition revoke active peer paths on clients, rather than relying only on ending server sessions. Advertise mesh as disabled while isolation applies, detect that change in a subsequent CONFIG even when tunnel settings are unchanged, and stop the client mesh state to wipe existing peer keys.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools



Remediation recommended

2. Packets left from offload mode go out with bad checksums 🐞 Bug ≡ Correctness
Description
The new skb_linearize() path in camex_chr_read() makes a leftover skb readable, but it does not
finish a CHECKSUM_PARTIAL checksum and does not split a GSO superpacket. camex_read_vnet()
passed both of those to userspace through the vnet header. Once CAMEX_IOC_VNET is turned off, a
queued TCP packet from the HW_CSUM/TSO period reaches the ordinary reader with an unfinished
transport checksum, and the peer drops it. A TSO superpacket that fits the read buffer is handed
over as one datagram larger than the MTU.
Code

src/camex-k.c[R672-675]

+    if (skb_linearize(skb)) {
+        dev_kfree_skb(skb);
+        return -ENOMEM;
+    }
Evidence
CAMEX_OFFLOADS includes NETIF_F_HW_CSUM and NETIF_F_TSO. camex_read_vnet() handles GSO and
CHECKSUM_PARTIAL skbs by putting NEEDS_CSUM and gso fields in the header for userspace to finish.
The new non-vnet path only linearises and then copies skb->data. skb_linearize changes neither
ip_summed nor the GSO state, so the checksum stays partial.

src/camex-k.c[442-460]
src/camex-k.c[660-683]

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
After CAMEX_IOC_VNET is turned off, `camex_chr_read()` linearises skbs queued while offloads were on. It does not compute a CHECKSUM_PARTIAL checksum or segment GSO packets, so userspace gets packets with bad checksums or packets larger than the MTU.
## Fix Focus Areas
- src/camex-k.c[665-675]
## Recommended Fix
On the non-vnet path, drop the skb if `skb_is_gso(skb)` is true and count it in `tx_dropped`. Alternatively, segment it with `skb_gso_segment`. If `skb->ip_summed == CHECKSUM_PARTIAL`, call `skb_checksum_help(skb)` after `skb_linearize`, and drop the skb if that fails.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools


3. Clients can lose server trust records ✓ Resolved
Description
client_known_store() proceeds with its read-and-rename rewrite even when opening the lock file
fails or all 40 flock() attempts fail. If another writer holds the lock for more than roughly two
seconds, both clients can publish snapshots based on stale contents, losing a server key or the
pake mark that prevents fallback to an unproven server.

4. Mesh clients can lose their IP on reload ✓ Resolved
Description
server_networks_apply() ends sessions after isolation is enabled by calling server_drop_entry(),
which does not retain an auto-client's address lease. When the network and its address pool are
otherwise unchanged, a reconnecting client no longer has the lease that server_end_client()
preserves for clients that may still be alive, so another client can take its address.

View review recommended (1)
5. An expired device removes a live neighbour ✓ Resolved
Description
remote_expire() deletes a neighbour by IP without checking whether the kernel entry still belongs
to that remote's MAC. If a newer remote has learned the same IP and replaced the neighbour entry, a
NAT lookup that encounters the older, expired record first deletes the newer remote's entry,
disrupting host-originated traffic until it is restored.


Informational

6. Turning isolation on disconnects clients that never had a direct path 🐞 Bug ≡ Correctness
Description
In server_networks_apply(), the new isolation branch ends every client with c->mesh set in a
network whose isolation has just come on. It never checks netw_table[n].mesh, but direct paths
only exist when the network allows mesh: server_mesh_control() and the relay accounting both
return early when netw_table[...].mesh is off. So a reload that turns isolation on in a network
with mesh off still disconnects every client that sent MESH=1, though none of them could have set
up a direct path.
Code

src/server.c[R5177-5183]

+        if (netw_table[n].active && c->mesh && net_isolation(n) &&
+            !netw_table[n].iso_applied) {
+            log_message(LOG_NOTICE, "Ended session of client %s: client "
+                        "isolation is on in its network now", c->client_id);
+            /* The same network: it keeps its address when it comes back */
+            server_end_client(c);
+            continue;
Evidence
Peer keys and direct paths are only handed out when both the client asked for mesh and the network
allows it (!e->mesh || ... || !netw_table[e->net1 - 1U].mesh returns early, and the CONFIG mesh
flag is netw_table[n].mesh && entry->mesh). The new condition tests only c->mesh, so clients in
a non-mesh network fall into the branch too.

src/server.c[3405-3414]
src/server.c[1290-1295]
src/server.c[5171-5184]

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
The isolation-on branch in `server_networks_apply()` ends every client with `c->mesh` set, even in networks where mesh is off. In those networks no direct path can exist, so the disconnect is unnecessary.
## Fix Focus Areas
- src/server.c[5177-5183]
## Recommended Fix
Add `netw_table[n].mesh &&` to the condition, so it reads `netw_table[n].active && netw_table[n].mesh && c->mesh && net_isolation(n) && !netw_table[n].iso_applied`.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools


7. A live TAP client ignores a server MTU equal to its default 🐞 Bug ≡ Correctness
Description
client_check_config_change() and client_config_same() compare the numeric mtu but ignore
mtu_given, even though that flag determines the MTU used when opening a TAP device. When a running
client has the default numeric MTU of 1420 without --mtu and the server later explicitly supplies
MTU=1420, both checks treat the CONFIG as unchanged, leaving the device at 1500 until the client
restarts.
Code

src/camex.c[2197]

+    current_config.mtu_given = next->mtu_given;
Evidence
A client without an explicit MTU retains 1420 internally, but packet_open() uses 1500 for its TAP
device when mtu_given is unset. Parsing a later server CONFIG with MTU=1420 sets mtu_given
without changing the numeric value; because both the incoming CONFIG comparison and
client_config_same() check only that value, neither detects the change needed to reopen the
device.

src/client.c[612-626]
src/camex.c[2127-2136]
src/config.c[1294-1297]
src/packet.c[86-90]
src/main.c[573-582]
src/packet.c[79-94]
src/proto.c[459-469]
src/config.c[1293-1297]
src/client.c[602-629]
src/camex.c[2127-2165]

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
The tunnel configuration change checks ignore `mtu_given`. A server CONFIG that explicitly supplies the client's existing numeric default MTU must reconfigure a live TAP device so it uses that MTU rather than 1500.
## Fix Focus Areas
- src/client.c[602-629]
- src/camex.c[2127-2165]
- src/config.c[1293-1297]
## Recommended Fix
Add `probe.mtu_given != current_config.mtu_given` to the condition in `client_check_config_change()` and `a->mtu_given == b->mtu_given` to `client_config_same()`. This makes a change in MTU explicitness trigger reconfiguration and reopen the TAP backend with the correct MTU.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools


8. Isolation branch can keep a lease for a replaced network 🐞 Bug ≡ Correctness
Description
server_networks_apply() takes the isolation-on branch before checking whether c->net_gen matches
netw_table[n].gen, sending a client from a replaced network through server_end_client() instead
of server_drop_entry(). When a newly isolated network reuses the old network’s slot and pool, the
client’s old address can remain leased to it, blocking other clients from using that address while
the log incorrectly says isolation was turned on for the client’s network.
Code

src/server.c[R5177-5183]

+        if (netw_table[n].active && c->mesh && net_isolation(n) &&
+            !netw_table[n].iso_applied) {
+            log_message(LOG_NOTICE, "Ended session of client %s: client "
+                        "isolation is on in its network now", c->client_id);
+            /* The same network: it keeps its address when it comes back */
+            server_end_client(c);
+            continue;
Evidence
netw_sync clears iso_applied when a slot changes hands and assigns the replacement network a new
generation, so the isolation condition can hold even though the client’s generation no longer
matches. The isolation branch runs before generation validation, and server_end_client() can
retain a lease when the old IP is in the replacement network’s pool; the comment that the client is
on “the same network” assumes a generation match that was never checked.

src/server.c[823-845]
src/netw.c[467-486]
src/netw.c[462-485]
src/server.c[5177-5194]
src/server.c[823-843]
src/server.c[391-397]
src/server.c[1184-1200]

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
The isolation-on branch can call `server_end_client()` for a client whose network slot was taken by another network, preserving the client’s old address as a lease in the replacement network’s pool.
## Fix Focus Areas
- src/server.c[5177-5194]
## Recommended Fix
Add `c->net_gen == netw_table[n].gen` to the isolation branch condition. Let clients with mismatched generations fall through to the existing `server_drop_entry()` path so they cannot create leases in the replacement network.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools


Grey Divider

Tip of the day
💡 Did you know, you can add REVIEW.md to your repo root and Qodo follows it on every PR

More tips ↗ | Customize Qodo ↗ | Qodo docs ↗

Grey Divider

Qodo Logo

…hared neighbour entries

- known_servers: when the lock cannot be taken (busy for 2 s, or not
  writable), the key is not kept this time instead of a rewrite that could
  drop another writer's line.
- Isolation turned on: the mesh client's session ends with
  server_end_client(), so its address stays leased to it.
- --bridge-mode nat: an expired remote removes its neighbour entry only
  when no live remote has that address.
@compot-dev

Copy link
Copy Markdown
Collaborator Author

/review

@compot-dev

Copy link
Copy Markdown
Collaborator Author

On the Qodo review of ea265ee, all three fixed in 224c126:

  1. known_servers without the lock: when the lock cannot be taken in 2 s (or the file's directory is not writable), the key is not kept this time (a warning; it is asked for again at the next start) instead of a rewrite that could drop another writer's line.
  2. Leases on isolation: the mesh client's session now ends with server_end_client(), which keeps its address leased to its ID; the network is the same, so it gets it back.
  3. Shared neighbour entries: an expired remote removes its neighbour entry only when no live remote has that address.

Checked locally: make test, e2e mesh, bridge, networks, hardening, reconfig, isolation, tap, key-binding, and Windows (x86_64, i686) under Wine.

@compot-dev

Copy link
Copy Markdown
Collaborator Author

/review

@compot-dev
compot-dev merged commit 5a490f1 into master Nov 21, 2026
18 checks passed
@compot-dev
compot-dev deleted the qodo-202-fixes branch November 21, 2026 23:42
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant