Improve --url-prefix handling with trailing slash normalization and redirect

- Normalize --url-prefix by stripping trailing slashes (so "/" becomes ""
  and "/llm/" becomes "/llm")
- Add send_redirect() method for HTTP redirects
- Redirect requests to prefix without trailing slash (e.g., /llm -> /llm/)
- Reject prefix collisions (e.g., /llmfoo when prefix is /llm)
- Update man page example to use /llm instead of /lamafiler

Author: vlasky

llamafile/flags.cpp

@@ -313,15 +313,17 @@ void llamafile_get_flags(int argc, char **argv) {
         if (!strcmp(flag, "--url-prefix")) {
             if (i == argc)
                 missing("--url-prefix");
-            FLAG_url_prefix = argv[i++];
+            // Normalize: strip trailing slashes in-place (so "/" becomes "" and "/llm/" becomes "/llm")
+            char *val = argv[i++];
+            size_t len = strlen(val);
+            while (len > 0 && val[len - 1] == '/') {
+                val[--len] = '\0';
+            }
+            FLAG_url_prefix = val;
             if (!IsAcceptablePath(FLAG_url_prefix, -1)) {
                 tinyprint(2, "error: --url-prefix must not have // or /. or /./ or /../\n", NULL);
                 exit(1);
             }
-            if (endswith(FLAG_url_prefix, "/")) {
-                tinyprint(2, "error: --url-prefix must not be slash or end with slash\n", NULL);
-                exit(1);
-            }
             continue;
         }
 

llamafile/server/client.cpp

@@ -323,6 +323,21 @@ Client::send_error(int code, const char* reason)
     return false;
 }
 
+// sends an http redirect response
+//
+// @param code must be a redirect status code (e.g. 301, 302, 307, 308)
+// @param location is the URL to redirect to
+bool
+Client::send_redirect(int code, const std::string_view& location)
+{
+    SLOG("redirect %d %.*s", code, (int)location.size(), location.data());
+    char* p = append_http_response_message(obuf_.p, code, nullptr);
+    p = stpcpy(p, "Location: ");
+    p = (char*)mempcpy(p, location.data(), location.size());
+    p = stpcpy(p, "\r\n");
+    return send_response(obuf_.p, p, "");
+}
+
 // appends start of http response message to `p`
 //
 // after this function is called, more header lines may be appended.
@@ -677,11 +692,22 @@ Client::dispatcher()
     std::string_view p1 = path();
     WRITE64LE(method, msg_.method);
     SLOG("%s %.*s", method, (int)p1.size(), p1.data());
-    if (!p1.starts_with(FLAG_url_prefix)) {
-        SLOG("path prefix mismatch");
-        return send_error(404);
+    size_t prefix_len = strlen(FLAG_url_prefix);
+    if (prefix_len > 0) {
+        if (!p1.starts_with(FLAG_url_prefix)) {
+            SLOG("path prefix mismatch");
+            return send_error(404);
+        }
+        // Exact match (e.g. /llm) -> redirect to /llm/
+        if (p1.size() == prefix_len)
+            return send_redirect(301, std::string(FLAG_url_prefix) + "/");
+        // Must have slash after prefix (e.g. /llm/foo not /llmfoo)
+        if (p1[prefix_len] != '/') {
+            SLOG("path prefix mismatch");
+            return send_error(404);
+        }
+        p1 = p1.substr(prefix_len);
     }
-    p1 = p1.substr(strlen(FLAG_url_prefix));
     if (!p1.starts_with("/") || !IsAcceptablePath(p1.data(), p1.size())) {
         SLOG("unacceptable path");
         return send_error(400);

llamafile/server/client.h

@@ -88,6 +88,7 @@ struct Client
     bool send_binary(const void*, size_t) __wur;
     void defer_cleanup(void (*)(void*), void*);
     bool send_error(int, const char* = nullptr);
+    bool send_redirect(int, const std::string_view&);
     char* append_http_response_message(char*, int, const char* = nullptr);
     bool send_response(char*, char*, const std::string_view) __wur;
     bool send_response_start(char*, char*) __wur;

llamafile/server/main.1

@@ -95,7 +95,7 @@ Hide system prompt from web user interface.
 Hide llamafile logo icon from web ui.
 .It Fl Fl url-prefix Ar URLPREFIX
 Specifies a URL prefix (subdirectory) under which the HTTP server will
-make the API accessible, e.g. /lamafiler. Useful when running llamafiler
+make the API accessible, e.g. /llm. Useful when running llamafiler
 behind a reverse proxy such as NGINX or Redbean. By default, this is set
 to / (root).
 .It Fl Fl verbose

llamafile/server/main.1.asc

@@ -103,9 +103,9 @@
 
        --url-prefix URLPREFIX
                Specifies a URL prefix  (subdirectory)  under  which  the  HTTP
-               server  will  make  the API accessible, e.g. /lamafiler. Useful
-               when running llamafiler behind a reverse proxy such as NGINX or
-               Redbean. By default, this is set to / (root).
+               server  will  make  the  API accessible, e.g. /llm. Useful when
+               running llamafiler behind a reverse proxy such as NGINX or Red‐
+               bean. By default, this is set to / (root).
 
        --verbose
                Enable logging of diagnostic information. This flag  is  useful