## Assistant Chat: HTTP 417 from `check_generation` treated as error instead of pending state

[lineumaciel]

Relates to: #3978

Description

The Nextcloud server intentionally returns HTTP 417 Expectation Failed from the check_generation endpoint as a signal that the AI task is still processing (pending). However, the new iOS Assistant Chat client (NCAssistantChatModel.swift, introduced in #3978) treats any 4xx response as a fatal error,

[lineumaciel]

The workaround for this issue in NGINX is to add a block that changes the response code from 417 to 202.

map $http_user_agent $ios_pending_response {
    default         0;
    ~*Nextcloud-iOS 1;
}

    location ~ ^/ocs/v2.php/apps/assistant/chat/check_generation {
        include /config/nginx/proxy.conf;
        include /config/nginx/resolver.conf;
        set $upstream_app nextcloud-server;
        set $upstream_port 80;
        set $upstream_proto http;
        proxy_pass $upstream_proto://$upstream_app:$upstream_port;

        proxy_hide_header Referrer-Policy;
        proxy_hide_header X-Content-Type-Options;
        proxy_hide_header X-Frame-Options;
        proxy_hide_header X-XSS-Protection;
        proxy_buffering off;

       
        if ($ios_pending_response = 1) {
            error_page 417 =202 @assistant_pending;
        }
    }

    location @assistant_pending {
        default_type application/json;
        return 202 '{"ocs":{"meta":{"status":"pending","statuscode":202,"message":"Task pending"},"data":{}}}';
    }