feat(github): implement PRReader + FileReader client (#80) #93

Closed
rodin wants to merge 16 commits from review-bot-issue-80 into feature/github-support
4 changed files with 120 additions and 3 deletions
Showing only changes of commit 1fcc0b738a - Show all commits
+31 -3
View File
14
@@ -133,10 +133,23 @@ func NewClient(token, baseURL string, opts ...ClientOption) *Client {
// SetHTTPClient sets the underlying HTTP client used for requests.
// This is intended for testing to inject mock transports.
// Passing nil will restore the default client with a 30s timeout.
// Passing nil restores the default client (30s timeout + auth-stripping
// CheckRedirect policy matching NewClient).
func (c *Client) SetHTTPClient(hc *http.Client) {
if hc == nil {
hc = &http.Client{Timeout: 30 * time.Second}
hc = &http.Client{
Timeout: 30 * time.Second,
CheckRedirect: func(req *http.Request, via []*http.Request) error {
if len(via) >= 10 {
return fmt.Errorf("stopped after 10 redirects")
}
Review

[MINOR] SetHTTPClient allows replacing the underlying http.Client without enforcing a safe CheckRedirect policy, which could cause Authorization to be forwarded to untrusted hosts on redirects. The docstring warns about this, but adding a runtime safeguard (e.g., rejecting clients with nil CheckRedirect or wrapping it) would further reduce misuse risk.

**[MINOR]** SetHTTPClient allows replacing the underlying http.Client without enforcing a safe CheckRedirect policy, which could cause Authorization to be forwarded to untrusted hosts on redirects. The docstring warns about this, but adding a runtime safeguard (e.g., rejecting clients with nil CheckRedirect or wrapping it) would further reduce misuse risk.
prev := via[len(via)-1]
if req.URL.Host != prev.URL.Host || (prev.URL.Scheme == "https" && req.URL.Scheme == "http") {
req.Header.Del("Authorization")
}
return nil
},
}
}
c.httpClient = hc
}
1
@@ -199,6 +212,9 @@ func (c *Client) doRequest(ctx context.Context, method, reqURL string, accept st
return nil, fmt.Errorf("create request: %w", err)
}
if c.token != "" {
// Bearer is the OAuth2 standard and is accepted by GitHub for both
// classic PATs and fine-grained tokens. The alternative "token" scheme
// is GitHub-specific and offers no additional compatibility.
req.Header.Set("Authorization", "Bearer "+c.token)
}
req.Header.Set("User-Agent", userAgent)
Review

[NIT] The timer pattern timer.Stop() // no-op after fire; kept for symmetry in the case <-timer.C branch is slightly misleading. Stop() after a channel receive is indeed a no-op since the channel has already fired, but the comment could be clearer: timer.Stop() after <-timer.C does nothing and could simply be omitted. This is a readability nit, not a correctness issue.

**[NIT]** The timer pattern `timer.Stop() // no-op after fire; kept for symmetry` in the `case <-timer.C` branch is slightly misleading. `Stop()` after a channel receive is indeed a no-op since the channel has already fired, but the comment could be clearer: `timer.Stop()` after `<-timer.C` does nothing and could simply be omitted. This is a readability nit, not a correctness issue.
7
@@ -232,7 +248,8 @@ func (c *Client) doRequest(ctx context.Context, method, reqURL string, accept st
// Retry on 429 rate limit
if resp.StatusCode == http.StatusTooManyRequests && attempt < maxAttempts-1 {
// Check for Retry-After header and override backoff if present
// Check for Retry-After header and override backoff if present.
// Supports both integer seconds (common) and HTTP-date format (RFC 7231).
if ra := resp.Header.Get("Retry-After"); ra != "" {
if seconds, err := strconv.Atoi(ra); err == nil && seconds > 0 {
delay := time.Duration(seconds) * time.Second
@@ -242,6 +259,17 @@ func (c *Client) doRequest(ctx context.Context, method, reqURL string, accept st
if attempt < len(backoff) {
backoff[attempt] = delay
}
} else if t, err := http.ParseTime(ra); err == nil {
delay := time.Until(t)
Review

[NIT] In doRequest, after the handleResponse call, there's a check if resp.StatusCode == http.StatusTooManyRequests but resp could theoretically have been closed by handleResponse. The body is closed, but the resp.StatusCode field is still accessible on the struct. This is correct and safe in Go's net/http — just worth being aware of.

**[NIT]** In `doRequest`, after the `handleResponse` call, there's a check `if resp.StatusCode == http.StatusTooManyRequests` but `resp` could theoretically have been closed by `handleResponse`. The body is closed, but the `resp.StatusCode` field is still accessible on the struct. This is correct and safe in Go's net/http — just worth being aware of.
if delay < 0 {
delay = 0
}
if delay > maxRetryAfter {
delay = maxRetryAfter
}
if attempt < len(backoff) {
backoff[attempt] = delay
}
}
}
continue
+81
View File
3
@@ -263,6 +263,84 @@ func TestDoRequest_RetryAfterDoesNotMutateBackoff(t *testing.T) {
}
}
func TestDoRequest_429RetryAfterHTTPDate(t *testing.T) {
attempts := 0
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
attempts++
if attempts == 1 {
// Use HTTP-date format (RFC 7231) — a time 2 seconds in the future.
future := time.Now().Add(2 * time.Second).UTC()
w.Header().Set("Retry-After", future.Format(http.TimeFormat))
w.WriteHeader(429)
w.Write([]byte(`{"message":"rate limit"}`))
return
}
w.WriteHeader(200)
w.Write([]byte(`{"ok":true}`))
}))
defer srv.Close()
c := NewClient("token", srv.URL, AllowInsecureHTTP())
c.SetHTTPClient(srv.Client())
c.SetRetryBackoff([]time.Duration{1 * time.Millisecond, 1 * time.Millisecond})
start := time.Now()
body, err := c.doGet(context.Background(), srv.URL+"/test")
elapsed := time.Since(start)
if err != nil {
t.Fatalf("unexpected error: %v", err)
}
if string(body) != `{"ok":true}` {
t.Errorf("unexpected body: %s", body)
}
if attempts != 2 {
t.Errorf("expected 2 attempts, got %d", attempts)
}
// HTTP-date was ~2s in the future; by the time client processes it,
// time.Until gives ~1-2s. Verify it's meaningfully delayed (not instant).
if elapsed < 500*time.Millisecond {
t.Errorf("expected meaningful delay from HTTP-date Retry-After, got %v", elapsed)
}
}
func TestDoRequest_429RetryAfterHTTPDateInPast(t *testing.T) {
attempts := 0
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
attempts++
if attempts == 1 {
// Use a time in the past — should result in zero/immediate retry.
past := time.Now().Add(-10 * time.Second).UTC()
w.Header().Set("Retry-After", past.Format(http.TimeFormat))
w.WriteHeader(429)
w.Write([]byte(`{"message":"rate limit"}`))
return
}
w.WriteHeader(200)
w.Write([]byte(`{"ok":true}`))
}))
defer srv.Close()
c := NewClient("token", srv.URL, AllowInsecureHTTP())
c.SetHTTPClient(srv.Client())
c.SetRetryBackoff([]time.Duration{5 * time.Second, 5 * time.Second})
start := time.Now()
_, err := c.doGet(context.Background(), srv.URL+"/test")
elapsed := time.Since(start)
if err != nil {
t.Fatalf("unexpected error: %v", err)
}
if attempts != 2 {
t.Errorf("expected 2 attempts, got %d", attempts)
}
// Past date should override the 5s backoff to ~0
if elapsed > 500*time.Millisecond {
t.Errorf("expected near-instant retry for past HTTP-date, got %v", elapsed)
}
}
func TestDoRequest_SetsUserAgentHeader(t *testing.T) {
var gotUA string
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
@@ -392,4 +470,7 @@ func TestSetHTTPClient_NilRestoresDefault(t *testing.T) {
if c.httpClient.Timeout != 30*time.Second {
t.Errorf("expected 30s timeout, got %v", c.httpClient.Timeout)
}
if c.httpClient.CheckRedirect == nil {
t.Fatal("expected CheckRedirect policy after SetHTTPClient(nil)")
}
}
+4
View File
4
@@ -22,6 +22,10 @@ func (c *Client) GetFileContent(ctx context.Context, owner, repo, path, ref stri
// If the path points to a single file (not a directory), the API returns
// a JSON object instead of an array; this is handled by returning a
// single-element slice.
//
// Note: dot-segments ("." and "..") in the path are silently removed to
// prevent path traversal. This means a path like "foo/../bar" resolves
// to "foo/bar" rather than "bar".
func (c *Client) ListContents(ctx context.Context, owner, repo, path string) ([]vcs.ContentEntry, error) {
reqURL := fmt.Sprintf("%s/repos/%s/%s/contents/%s",
c.baseURL, url.PathEscape(owner), url.PathEscape(repo), escapePath(path))
3
+4
View File
9
@@ -123,6 +123,10 @@ func (c *Client) GetPullRequestFiles(ctx context.Context, owner, repo string, nu
// GetFileContentAtRef fetches a file at a specific ref from a repo.
// If ref is empty, the query parameter is omitted (uses default branch).
//
// Note: dot-segments ("." and "..") in the path are silently removed to
// prevent path traversal. This means a path like "foo/../bar" resolves
// to "foo/bar" rather than "bar".
func (c *Client) GetFileContentAtRef(ctx context.Context, owner, repo, path, ref string) (string, error) {
reqURL := fmt.Sprintf("%s/repos/%s/%s/contents/%s",
Review

[NIT] The GetPullRequestFiles comment says 'Returns nil (not an empty slice) when the PR has no changed files' but if the first page returns an empty array, allFiles remains nil — this is correct. However the GitHub API actually returns an empty array for PRs with 0 changed files (valid edge case), so the nil vs empty distinction in the doc comment is accurate but subtle. No code change needed.

**[NIT]** The `GetPullRequestFiles` comment says 'Returns nil (not an empty slice) when the PR has no changed files' but if the first page returns an empty array, `allFiles` remains nil — this is correct. However the GitHub API actually returns an empty array for PRs with 0 changed files (valid edge case), so the nil vs empty distinction in the doc comment is accurate but subtle. No code change needed.
c.baseURL, url.PathEscape(owner), url.PathEscape(repo), escapePath(path))
12