Save notes whose body is empty

[dangduc]

Problem

Saving a note with an empty body failed with a modal alert:

Changed notes could not be saved because a parameter was invalid.

That text is OSStatus -50 (paramErr), raised by -noteDidNotWrite:errorCode:. Each failure also orphaned a zero-length temporary file in the notes directory.

Cause

A note with an empty body is stored as a zero-length file, and Foundation returns a NULL buffer pointer for zero-length NSData:

read 0-byte file       len=0 bytes=0x0
[data copy] (baseline) len=0 bytes=0x0
dataUsingEncoding      len=0 bytes=0x600000f1c000

Once such a note was read back from disk, -rememberSourceBaselineData:encoding: cached that NULL-pointer data. On the next save, the unchanged-source fast path in -sourceDataReturningError: returned it verbatim instead of re-encoding, and FSRefWriteData rejected it:

if (!buffer || !fsRef) { return paramErr; }

Traced directly:

begin name=vô lễ.txt nameLength=9 dataLength=0 dataBytes=0x0 destRef=.../vô lễ.txt
CreateTemporaryFile ok temp=.../.017801465-810886610-13
FSRefWriteData failed: -50 (dataLength=0 dataBytes=0x0 blockSize=1048576)
error writing to temporary file: -50
Unable to save note file vô lễ.txt

The temporary file is created before the write, so a failure leaves it behind — which is the signature these failures leave in a notes directory.

Two conditions hid this. A note created in the current session has no cached source baseline, so it re-encodes and saves fine; the failure needs a relaunch first. And it affects every empty-body note at once during any pass that rewrites all notes, such as a storage-format change, rather than a single deliberate save.

The earlier write path always built data with dataUsingEncoding:, which returns a non-NULL pointer for an empty string, so this arrived with the source-editing work in 60e0333.

Fix

• Sources/Utilities/BufferUtils.c — a zero-length write is valid, so only a missing buffer with bytes to write is an error. This also covers the export path, which passes [formattedData bytes] and had the same latent failure.
• Sources/Model/NoteObject.m — skip the unchanged-source fast path when the cached data is empty, so a NULL pointer never reaches a writer.

Testing

New suite Tests/Regression/empty-source-write/, registered in Tests/run-regression-tests.py (18 checks):

• zero-length data exposes a NULL buffer pointer, and a zero-length file reads back that way
• FSRefWriteData accepts a zero-length write with a NULL buffer, and still rejects a NULL buffer with bytes to write, and a missing FSRef
• an empty-body note writes on creation and rewrites after its source baseline is cached
• no orphaned temporary files remain afterwards

Verified against a copy of a real library with five empty-body notes: 5 failures and 5 orphaned temporary files before, 0 and 0 after. Reverting either guard and rebuilding fails the suite (FAIL: zero-length write with a NULL buffer succeeds), so the check cannot pass vacuously.

Ran on macOS 13.7.8, Intel Development build.

https://claude.ai/code/session_01W5dVNP811hKcLKUmJHD1FG