[codex] fix(vscode): fall back to git CLI for diffs

[yzlu0917]

Summary

• keep using the VS Code Git extension as the primary source for diffs
• fall back to git diff / git diff --cached from workspace or repo roots when the Git API returns no diff content
• add regression coverage for splitting combined diffs and CLI fallback behavior

Why

In some remote environments such as WSL, Continue can end up with an empty diff from the VS Code Git API even though git diff in the integrated terminal shows changes. That causes the @Git Diff context provider to incorrectly say Git shows no current changes.

This change preserves the existing fast path but adds a CLI fallback so diff context still works when the Git extension cannot provide the patch content.

Validation

• ran npx vitest run src/util/gitDiffFallback.vitest.ts in extensions/vscode
• ran git diff --check

Closes #9136

---

Summary by cubic

Add a CLI fallback for Git diffs in VS Code so the @Git Diff context works when the Git API returns no content (e.g., WSL/remote). Also adds dynamic model fetching across providers with UI support to list available models by API key.

• Bug Fixes

  • VS Code: fall back to git diff --cached and git diff from repo/workspace roots discovered via git rev-parse --show-toplevel (deduped, skip non-git, 10MB buffer).
  • Split combined patch output into per-file entries with splitGitDiff; return only non-empty diffs.
  • Use the CLI fallback in ideUtils when API diffs are empty; added vitest coverage for split logic and fallback behavior.
  • Fixes @Git Diff context provider fails in WSL (Remote) - "Git shows no current changes" #9136.

• New Features

  • Core/Webview: added "models/fetch" protocol and fetchModels for ollama, openrouter, anthropic, gemini, and generic providers via listModels; surfaces errors via IDE toast.
  • GUI: Add Model screen can fetch provider models with the entered API key; initializes dynamic Ollama/OpenRouter catalogs; adds GPT‑5.4 presets and Gemma 4; minor listbox label tweaks.
  • LLMs/Tools: broadened tool-support detection (gpt‑4+ and o‑series), OpenAI responses gating via isOSeriesOrGpt5PlusModel; improved Gemma image support regex; OpenRouter sets default OPENROUTER_HEADERS from @continuedev/openai-adapters.

^{Written for commit fa2ed25. Summary will update on new commits.}

[cubic-dev-ai]

1 issue found across 3 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="extensions/vscode/src/util/gitDiffFallback.ts">

<violation number="1" location="extensions/vscode/src/util/gitDiffFallback.ts:12">
P2: Diff splitting regex is not line-anchored, so hunk content containing `diff --git ...` can be misinterpreted as a new file header and corrupt split results.</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review, or fix all with cubic.}

[cubic-dev-ai]

P2: Diff splitting regex is not line-anchored, so hunk content containing diff --git ... can be misinterpreted as a new file header and corrupt split results.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At extensions/vscode/src/util/gitDiffFallback.ts, line 12:

<comment>Diff splitting regex is not line-anchored, so hunk content containing `diff --git ...` can be misinterpreted as a new file header and corrupt split results.</comment>

<file context>
@@ -0,0 +1,69 @@
+const GIT_DIFF_MAX_BUFFER = 10 * 1024 * 1024;
+
+export function splitGitDiff(diffString: string): string[] {
+  const fileDiffHeaderRegex = /(?=diff --git a\/.* b\/.*)/;
+  const diffs = diffString.split(fileDiffHeaderRegex);
+
</file context>

Suggested change

	const fileDiffHeaderRegex = /(?=diff --git a\/.* b\/.*)/;
	const fileDiffHeaderRegex = /(?=^diff --git a\/.* b\/.*$)/m;

[chatgpt-codex-connector]

💡 Codex Review

continue/gui/src/forms/AddModelForm.tsx

Lines 75 to 76 in e030a32

	setFetchedModelsList((prev) =>
	selectedProvider.provider === providerAtFetchTime ? models : prev,

Keep async model fetches scoped to the current provider

The guard in setFetchedModelsList compares selectedProvider.provider to providerAtFetchTime, but both values are captured from the same render, so this condition is effectively always true for that request. If a user starts fetching models, switches providers before the request resolves, and the old request returns later, the old provider's model list can overwrite the current provider's options, leading to mismatched selections and incorrect model/provider combinations being saved.

---

continue/core/llm/fetchModels.ts

Lines 180 to 181 in e030a32

	const base = apiBase || "https://generativelanguage.googleapis.com/v1beta/";
	const url = new URL("models", base);

Preserve Gemini API base path when composing /models URL

Constructing the endpoint with new URL("models", base) breaks for user-provided Gemini apiBase values that omit a trailing slash (for example, .../v1beta resolves to .../models instead of .../v1beta/models). In that common input form, model fetching hits the wrong endpoint and fails, so the Add Model flow cannot discover Gemini models even with a valid key/base pair.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".