fix: prototype of visual buf replacement on demand is workingHEADmaster

1 files changed, 80 insertions, 67 deletions

diff --git a/lua/99/providers.lua b/lua/99/providers.lua
index 8875ca8..63e64aa 100644
--- a/lua/99/providers.lua
+++ b/lua/99/providers.lua
@@ -1,7 +1,7 @@
 --- @class _99.Providers.Observer
 --- @field on_stdout fun(line: string): nil
 --- @field on_stderr fun(line: string): nil
---- @field on_complete fun(status: _99.Prompt.EndingState, res: string): nil
+--- @field on_complete fun(status: _99.Prompt.EndingState, res: string|table): nil
 --- @field on_start fun(): nil
 
 --- @param fn fun(...: any): nil
@@ -62,54 +62,6 @@ end
 --- @field predictedPerTkMs number
 --- @field predictedPerSec number
 
---- @param raw table
---- @return _A4.OpenAIResponse
-local function parse_openai_response(raw)
-    local choice = raw.choices[1]
-
-    --- @type _A4.OpenAIResponse
-    local response = {
-        completion = {
-            finishReason = choice.finish_reason,
-            index = choice.index,
-            message = {
-                role = choice.message.role,
-                content = choice.message.content,
-            },
-        },
-        metadata = {
-            created = raw.created,
-            model = raw.model,
-            systemFingerPrint = raw.system_fingerprint,
-            object = raw.object,
-            usage = {
-                completionTks = raw.usage.completion_tokens,
-                promptTks = raw.usage.prompt_tokens,
-                totalTks = raw.usage.total_tokens,
-                promptTksDetails = {
-                    cachedTks = raw.usage.prompt_tokens_details.cached_tokens,
-                },
-            },
-        },
-        id = {
-            id = raw.id,
-        },
-        timings = {
-            cacheN = raw.timings.cache_n,
-            promptN = raw.timings.prompt_n,
-            promptMs = raw.timings.prompt_ms,
-            promptPerTkMs = raw.timings.prompt_per_token_ms,
-            promptPerSec = raw.timings.prompt_per_second,
-            predictedN = raw.timings.predicted_n,
-            predictedMs = raw.timings.predicted_ms,
-            predictedPerTkMs = raw.timings.predicted_per_token_ms,
-            predictedPerSec = raw.timings.predicted_per_second,
-        },
-    }
-
-    return response
-end
-
 --- @class _99.Providers.BaseProvider
 --- @field _build_command fun(self: _99.Providers.BaseProvider, query: string, context: _99.Prompt): string[]
 --- @field _get_provider_name fun(self: _99.Providers.BaseProvider): string
@@ -123,8 +75,8 @@ end
 
 --- @param data string
 --- @return string
-function BaseProvider:_clean_response(data)
-    local _data, _ = data:gsub("```[a-z]*\n", ""):gsub(" ```", "")
+function BaseProvider._clean_response(data)
+    local _data, _ = data:gsub("```[a-z]*\n", ""):gsub("```", "")
     return _data
 end
 
@@ -164,9 +116,6 @@ function BaseProvider:_retrieve_response(context)
     end
 end
 
--- TODO: Remember that we are ditching the tmp_file; likely 90%
--- You can just grab the json data and dump it in logs honestly
--- files only make sense for agentic flows
 --- @param query string
 --- @param context _99.Prompt
 --- @param observer _99.Providers.Observer
@@ -192,21 +141,25 @@ function BaseProvider:make_request(query, context, observer)
         command,
         { text = true },
         vim.schedule_wrap(function(obj)
-            if obj.code ~= 0 then
-                once_complete("failed", obj.stderr)
+            logger:debug("exit callback fired", "code", tostring(obj.code))
+            logger:debug("schedule_wrap", "stdout", tostring(obj.stdout))
+            logger:debug("schedule_wrap", "stderr", tostring(obj.stderr))
+
+            if context:is_cancelled() then
+                once_complete("cancelled", "")
                 return
             end
 
-            local data = parse_openai_response(vim.json.decode(obj.stdout))
-            if self._get_provider_name == "BareMetalProvider" then
-                data.completion.message.content =
-                    self:_clean_response(data.completion.message.content)
-                context.req_response = data.completion.message.content
+            if obj.code ~= 0 then
+                once_complete("failed", obj.stderr or "")
+                return
             end
-            once_complete("success", data.completion.message.content)
+
+            once_complete("success", obj.stdout)
         end)
     )
 
+    logger:debug("proc spawned", "proc_id", tostring(proc))
     context:_set_process(proc)
 end
 
@@ -400,7 +353,8 @@ function GeminiCLIProvider._get_default_model()
 end
 
 --- @class BareMetalProvider : _99.Providers.BaseProvider
---- @field _clean_response fun(self: _99.Providers.BaseProvider, data : string): string
+--- @field _clean_response fun(data : string): string
+--- @field _parse_openai_response fun(raw: table): _A4.OpenAIResponse
 local BareMetalProvider = setmetatable({}, { __index = BaseProvider })
 
 --- @param query string
@@ -412,11 +366,22 @@ function BareMetalProvider._build_command(_, query, context)
         "-s",
         context.endpoint,
         "-H",
-        '"Content-Type: applications/json"',
+        "Content-Type: application/json",
         "-d",
-        '"{"messages":[{"role":"system","content":"You are a strict code completion backend for Neovim. Your input will be a code snippet, function signature, or a comment requesting code. CRITICAL DIRECTIONS: 1. Output ONLY valid, executable programming code. 2. Do NOT wrap your response in markdown code blocks. 3. Do NOT include any conversational filler, explanations, greetings, or sign-offs. 4. If you cannot fulfill the request, output nothing or a code comment explaining why."},{"role":"user","content":"'
-            .. query
-            .. '"}],"temperature":0.0,"stream":false}"',
+        vim.json.encode({
+            messages = {
+                {
+                    role = "system",
+                    content = "You are a strict code completion backend for Neovim. Your input will be a code snippet, function signature, or a comment requesting code. CRITICAL DIRECTIONS: 1. Output ONLY valid, executable programming code. 2. Do NOT wrap your response in markdown code blocks. 3. Do NOT include any conversational filler, explanations, greetings, or sign-offs. 4. If you cannot fulfill the request, output nothing or a code comment explaining why.",
+                },
+                {
+                    role = "user",
+                    content = query,
+                },
+            },
+            temperature = 0.0,
+            stream = false,
+        }),
     }
 end
 
@@ -436,6 +401,54 @@ function BareMetalProvider.fetch_models(callback)
     }, nil)
 end
 
+--- @param raw table
+--- @return _A4.OpenAIResponse
+function BareMetalProvider._parse_openai_response(raw)
+    local choice = raw.choices[1]
+
+    --- @type _A4.OpenAIResponse
+    local response = {
+        completion = {
+            finishReason = choice.finish_reason,
+            index = choice.index,
+            message = {
+                role = choice.message.role,
+                content = choice.message.content,
+            },
+        },
+        metadata = {
+            created = raw.created,
+            model = raw.model,
+            systemFingerPrint = raw.system_fingerprint,
+            object = raw.object,
+            usage = {
+                completionTks = raw.usage.completion_tokens,
+                promptTks = raw.usage.prompt_tokens,
+                totalTks = raw.usage.total_tokens,
+                promptTksDetails = {
+                    cachedTks = raw.usage.prompt_tokens_details.cached_tokens,
+                },
+            },
+        },
+        id = {
+            id = raw.id,
+        },
+        timings = {
+            cacheN = raw.timings.cache_n,
+            promptN = raw.timings.prompt_n,
+            promptMs = raw.timings.prompt_ms,
+            promptPerTkMs = raw.timings.prompt_per_token_ms,
+            promptPerSec = raw.timings.prompt_per_second,
+            predictedN = raw.timings.predicted_n,
+            predictedMs = raw.timings.predicted_ms,
+            predictedPerTkMs = raw.timings.predicted_per_token_ms,
+            predictedPerSec = raw.timings.predicted_per_second,
+        },
+    }
+
+    return response
+end
+
 return {
     BaseProvider = BaseProvider,
     OpenCodeProvider = OpenCodeProvider,