Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.47%. Comparing base (8c57717) to head (76e5560).
⚠️ Report is 2 commits behind head on master.

@@           Coverage Diff           @@
##           master   #19275   +/-   ##
=======================================
  Coverage   98.47%   98.47%           
=======================================
  Files         176      176           
  Lines       22845    22862   +17     
=======================================
+ Hits        22497    22514   +17     
  Misses        348      348           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

Code size report:

Reference:  esp32: Make a generic sdkconfig.flash_qio_80m config snippet. [4fd7295]
Comparison: extmod/vfs_posix: Add size field to ilistdir() entries. [merge of 76e5560]
  mpy-cross:    +0 +0.000% 
   bare-arm:    +0 +0.000% 
minimal x86:    +0 +0.000% 
   unix x64:  +240 +0.028% standard
      stm32:    +0 +0.000% PYBV10
      esp32:    +0 +0.000% ESP32_GENERIC
     mimxrt:    +0 +0.000% TEENSY40
        rp2:    +0 +0.000% RPI_PICO_W
       samd:    +0 +0.000% ADAFRUIT_ITSYBITSY_M4_EXPRESS
  qemu rv32:    +0 +0.000% VIRT_RV32

Thanks for the contribution.

The main downside of this change is the performance impact: each iteration through ilistdir() result now needs to (1) allocate a long string for the full path; (2) call stat which on Linux goes through to the kernel. This penalty is paid for everyone using ilistdir on POSIX, regardless of whether they need the size or not.

Might be worth measuring the actual performance impact of this change on a large directory listing, to see how significant it is.

Good point, I quickly benchmarked it on x86-linux: its visible

• At 100,000 entries on ext4, the patch adds ~64 ms of wall time per listing (39.7 ms → 103.5 ms).
• Per-entry cost is roughly flat at ~0.4-0.8 μs across all N, which confirms the expected behaviour: one extra stat() syscall + one vstr append per entry.
• Throughput drops from ~2,500 entries/ms to ~970 entries/ms on ext4.

data:

┌─────────┬─────────┬──────────┬─────────┬───────────┬──────────┐
│ entries │ master  │ patched  │  added  │ per-entry │ slowdown │
├─────────┼─────────┼──────────┼─────────┼───────────┼──────────┤
│     100 │   51 μs │    88 μs │   37 μs │   0.37 μs │    1.73x │
├─────────┼─────────┼──────────┼─────────┼───────────┼──────────┤
│   1,000 │  457 μs │   878 μs │  421 μs │   0.42 μs │    1.92x │
├─────────┼─────────┼──────────┼─────────┼───────────┼──────────┤
│  10,000 │ 4.25 ms │  12.6 ms │  8.3 ms │   0.83 μs │    2.96x │
├─────────┼─────────┼──────────┼─────────┼───────────┼──────────┤
│  50,000 │ 20.5 ms │  48.7 ms │ 28.2 ms │   0.56 μs │    2.37x │
├─────────┼─────────┼──────────┼─────────┼───────────┼──────────┤
│ 100,000 │ 39.7 ms │ 103.5 ms │ 63.8 ms │   0.64 μs │    2.61x │
└─────────┴─────────┴──────────┴─────────┴───────────┴──────────┘

Plots (tempfs and ext4):

• tmpfs and ext4 land within noise of each other at these scales because the inode cache is hot; the syscall overhead dominates, not the disk read.

Options to mitigate that perf hit

A. fstatat(dirfd(self->dir), fn, ...) instead of stat(full_path, ...) - maybe a bit faster
B. Config gate (MICROPY_VFS_POSIX_ILISTDIR_SIZE, default off) - opt-in to 4 tuple
C. Return -1 for size (per docs: "size of the file or -1 if unknown") - feels a bit off tho
D. Move the fix to mpremote only -> Cost paid only by consumers who actually need size.
E. Keep status quo

Not sure