From 983e523ca9f8e6a7db56b4aca23a962be95e7051 Mon Sep 17 00:00:00 2001
From: t <t@tjp.lol>
Date: Mon, 6 Jul 2026 08:34:35 -0600
Subject: rendering & chat fixes for panto cli

- resiliency for tool ids that come back from some openai_chat providers
- fixing codepoint width recognition
- implement proper rendering for full markdown as seen from ai models
---
 libpanto/src/provider_openai_chat.zig | 84 ++++++++++++++++++++++++++++++++---
 1 file changed, 78 insertions(+), 6 deletions(-)

(limited to 'libpanto/src')

diff --git a/libpanto/src/provider_openai_chat.zig b/libpanto/src/provider_openai_chat.zig
index a05c7ba..4ac1e11 100644
--- a/libpanto/src/provider_openai_chat.zig
+++ b/libpanto/src/provider_openai_chat.zig
@@ -466,12 +466,12 @@ const StreamState = struct {
         }
         const tu = &self.active_tool.?;
 
-        // Append identity fragments. Most providers send id+name whole on
-        // the first delta and never repeat them, but appending is the only
-        // correct behavior across the full range of OpenAI-compatible
-        // backends — some chunk these strings.
-        if (d.id) |s| try tu.id_buf.appendSlice(self.allocator, s);
-        if (d.name) |s| try tu.name_buf.appendSlice(self.allocator, s);
+        // Some OpenAI-compatible backends send identity as fragments
+        // (`call_` then `xyz`), while others resend the cumulative/full value
+        // on later argument chunks. Accept both without turning repeats into
+        // runaway ids like `call_xyzcall_xyz...`.
+        if (d.id) |s| try mergeStreamingField(self.allocator, &tu.id_buf, s);
+        if (d.name) |s| try mergeStreamingField(self.allocator, &tu.name_buf, s);
 
         // Defer `onBlockStart` until args begin. The first argument
         // fragment is our signal that identity is likely settled enough
@@ -601,6 +601,24 @@ const StreamState = struct {
         try out.push(.{ .block_start = .{ .block_type = .ToolUse, .index = tu.block_index } });
     }
 
+    fn mergeStreamingField(
+        allocator: Allocator,
+        buf: *conversation.TextualBlock,
+        piece: []const u8,
+    ) !void {
+        if (piece.len == 0) return;
+        if (buf.items.len == 0) {
+            try buf.appendSlice(allocator, piece);
+        } else if (std.mem.eql(u8, buf.items, piece)) {
+            return;
+        } else if (std.mem.startsWith(u8, piece, buf.items)) {
+            buf.clearRetainingCapacity();
+            try buf.appendSlice(allocator, piece);
+        } else {
+            try buf.appendSlice(allocator, piece);
+        }
+    }
+
     /// End the stream: close any open text/thinking block, close the still-
     /// active tool_use (if any), then commit the assembled assistant
     /// Message to the conversation and push the terminal `message_complete`.
@@ -988,6 +1006,60 @@ test "fragmented tool_call id and name are reassembled" {
     try testing.expectEqualStrings("{\"host\":\"a.com\"}", tu.input.items);
 }
 
+test "repeated full tool_call id and name do not accumulate" {
+    const allocator = testing.allocator;
+
+    var conv = conversation.Conversation.init(allocator);
+    defer conv.deinit();
+    try addUserText(&conv, "read it");
+
+    const events = [_][]const u8{
+        \\{"choices":[{"delta":{"role":"assistant"}}]}
+        ,
+        \\{"choices":[{"delta":{"tool_calls":[{"index":0,"id":"chatcmpl-tool-abc","type":"function","function":{"name":"std__read","arguments":"{\"path\""}}]}}]}
+        ,
+        \\{"choices":[{"delta":{"tool_calls":[{"index":0,"id":"chatcmpl-tool-abc","function":{"name":"std__read","arguments":":\"a\"}"}}]}}]}
+        ,
+        \\{"choices":[{"delta":{},"finish_reason":"tool_calls"}]}
+        ,
+        "[DONE]",
+    };
+
+    try runStreamedTurn(allocator, &conv, null, &events);
+
+    const tu = conv.messages.items[1].content.items[0].ToolUse;
+    try testing.expectEqualStrings("chatcmpl-tool-abc", tu.id);
+    try testing.expectEqualStrings("std.read", tu.name);
+    try testing.expectEqualStrings("{\"path\":\"a\"}", tu.input.items);
+}
+
+test "cumulative tool_call id and name replace their prefix" {
+    const allocator = testing.allocator;
+
+    var conv = conversation.Conversation.init(allocator);
+    defer conv.deinit();
+    try addUserText(&conv, "read it");
+
+    const events = [_][]const u8{
+        \\{"choices":[{"delta":{"role":"assistant"}}]}
+        ,
+        \\{"choices":[{"delta":{"tool_calls":[{"index":0,"id":"chatcmpl-tool-","type":"function","function":{"name":"std_","arguments":"{\"path\""}}]}}]}
+        ,
+        \\{"choices":[{"delta":{"tool_calls":[{"index":0,"id":"chatcmpl-tool-abc","function":{"name":"std__read","arguments":":\"a\"}"}}]}}]}
+        ,
+        \\{"choices":[{"delta":{},"finish_reason":"tool_calls"}]}
+        ,
+        "[DONE]",
+    };
+
+    try runStreamedTurn(allocator, &conv, null, &events);
+
+    const tu = conv.messages.items[1].content.items[0].ToolUse;
+    try testing.expectEqualStrings("chatcmpl-tool-abc", tu.id);
+    try testing.expectEqualStrings("std.read", tu.name);
+    try testing.expectEqualStrings("{\"path\":\"a\"}", tu.input.items);
+}
+
 test "inbound wire tool name is decoded to dotted form (even split across __)" {
     // The model echoes the wire name it was given (`std__read`). It is
     // decoded to the internal `std.read` for the conversation/session/
-- 
cgit v1.3