espflasher: add retry logic for flash block writes and fix compressed timeouts

go.mod

@@ -5,12 +5,12 @@ go 1.22
 require (
 	github.com/stretchr/testify v1.7.0
 	go.bug.st/serial v1.6.2
+	golang.org/x/sys v0.0.0-20220829200755-d48e67d00261
 )
 
 require (
 	github.com/creack/goselect v0.1.2 // indirect
 	github.com/davecgh/go-spew v1.1.0 // indirect
 	github.com/pmezard/go-difflib v1.0.0 // indirect
-	golang.org/x/sys v0.0.0-20220829200755-d48e67d00261 // indirect
 	gopkg.in/yaml.v3 v3.0.0-20200313102051-9f266ea9e77c // indirect
 )

pkg/espflasher/flasher.go

@@ -737,8 +737,10 @@ func (f *Flasher) flashCompressed(data []byte, offset uint32, progress ProgressF
 		blockLen := min(compSize-sent, writeSize)
 
 		block := compressed[sent : sent+blockLen]
-		if err := f.conn.flashDeflData(block, seq); err != nil {
-			return fmt.Errorf("flash block %d of %d: %w", seq, numBlocks, err)
+		if err := f.retryFlashBlock(seq, numBlocks, func() error {
+			return f.conn.flashDeflData(block, seq)
+		}); err != nil {
+			return err
 		}
 
 		sent += blockLen
@@ -810,8 +812,10 @@ func (f *Flasher) flashUncompressed(data []byte, offset uint32, progress Progres
 			block = padded
 		}
 
-		if err := f.conn.flashData(block, seq); err != nil {
-			return fmt.Errorf("flash block %d of %d: %w", seq, numBlocks, err)
+		if err := f.retryFlashBlock(seq, numBlocks, func() error {
+			return f.conn.flashData(block, seq)
+		}); err != nil {
+			return err
 		}
 
 		sent += blockLen
@@ -923,6 +927,13 @@ func (f *Flasher) EraseRegion(offset, size uint32, progress ProgressFunc) error
 	})
 }
 
+// flashBlockRetries is the number of attempts for each flash data block write.
+// Transient serial errors (SLIP timeouts, framing glitches, USB CDC buffer
+// drops) are common during flashing and typically resolve on retry. The stub
+// handles duplicate sequence numbers gracefully, so resending an
+// already-processed block is safe.
+const flashBlockRetries = 3
+
 // eraseProgressInterval is the tick interval used by tickErase when reporting
 // synthetic erase progress via EraseFlash and EraseRegion.
 const eraseProgressInterval = 500 * time.Millisecond
@@ -1269,6 +1280,33 @@ func compressData(data []byte) ([]byte, error) {
 	return buf.Bytes(), nil
 }
 
+// retryFlashBlock attempts a flash block write up to flashBlockRetries times.
+// On failure, it flushes stale serial data and waits briefly before retrying.
+// This handles transient SLIP timeouts and framing errors that are common
+// during ESP32 flash operations, matching esptool.py's WRITE_BLOCK_ATTEMPTS
+// retry behavior.
+func (f *Flasher) retryFlashBlock(seq, numBlocks uint32, writeFn func() error) error {
+	var err error
+	for attempt := range flashBlockRetries {
+		err = writeFn()
+		if err == nil {
+			return nil
+		}
+		if attempt < flashBlockRetries-1 {
+			f.logf("Warning: block %d/%d write failed (attempt %d/%d): %v — retrying",
+				seq, numBlocks, attempt+1, flashBlockRetries, err)
+			// Wait before flushing so a delayed response from the timed-out
+			// attempt has time to arrive and gets cleared by the flush.
+			// Without this ordering, the stale response arrives after the
+			// flush and corrupts the retry's response parsing (e.g. SLIP
+			// END 0xC0 read as a status byte).
+			time.Sleep(200 * time.Millisecond)
+			f.conn.flushInput()
+		}
+	}
+	return fmt.Errorf("flash block %d of %d: %w", seq, numBlocks, err)
+}
+
 // logf logs a message if a logger is configured.
 func (f *Flasher) logf(format string, args ...interface{}) {
 	if f.opts.Logger != nil {

pkg/espflasher/protocol.go

@@ -92,6 +92,10 @@ type conn struct {
 	// deflCompSize is the compressed size passed to the most recent
 	// flashDeflBegin call, used to scale the flashDeflEnd ack timeout.
 	deflCompSize uint32
+	// deflUncompSize is the uncompressed size passed to the most recent
+	// flashDeflBegin call, used with deflCompSize to estimate the
+	// decompression ratio for per-block timeouts in flashDeflData.
+	deflUncompSize uint32
 }
 
 // isStub returns whether the stub loader is running.
@@ -456,9 +460,11 @@ func (c *conn) flashDeflBegin(uncompSize, compSize, offset uint32, encrypted boo
 
 	timeout := eraseTimeoutForSize(uncompSize)
 
-	// Remember the compressed size for this download so flashDeflEnd can
-	// scale its own ack timeout the same way.
+	// Remember sizes for this download so flashDeflData can estimate per-block
+	// decompressed sizes for timeout scaling, and flashDeflEnd can scale its
+	// own ack timeout.
 	c.deflCompSize = compSize
+	c.deflUncompSize = uncompSize
 
 	// ESP32-S2 and newer ROM bootloaders support a 5th parameter (encrypted
 	// flag). ESP8266 and original ESP32 ROM only accept 4 parameters (16 bytes).
@@ -488,7 +494,16 @@ func (c *conn) flashDeflData(block []byte, seq uint32) error {
 	binary.LittleEndian.PutUint32(data[12:16], 0)
 	copy(data[16:], block)
 
-	timeout := flashWriteTimeoutForSize(uint32(len(block)))
+	// The stub must decompress this block and write the result to flash
+	// before ACKing. The decompressed data can be much larger than the
+	// compressed block, so scale the timeout by the compression ratio
+	// to account for the actual flash write time.
+	estimatedSize := uint32(len(block))
+	if c.deflCompSize > 0 && c.deflUncompSize > c.deflCompSize {
+		ratio := float64(c.deflUncompSize) / float64(c.deflCompSize)
+		estimatedSize = uint32(float64(len(block)) * ratio)
+	}
+	timeout := flashWriteTimeoutForSize(estimatedSize)
 	_, err := c.checkCommand("write compressed flash block", cmdFlashDeflData, data, checksum(block), timeout, 0)
 	return err
 }