fix: report actual st_blocks for directories (XFS inline dirs)

[shivampwm2020]

Problem

gdu charges a directory a fixed one-block (4096 B) of disk usage instead of reading what the filesystem actually allocated to it.

On XFS a small directory is stored inline in the inode: stat reports Blocks: 0, xfs_bmap reports no extents, and du reports 0. gdu reported 4.0 KiB for such a directory (issue #393).

Root cause

gdu never read a directory's own st_blocks.

• Before fix: correct item count and size for empty dirs #574 the analyzers hardcoded totalSize/totalUsage = 4096 for every directory, which is exactly the reported 4.0 KiB on an inline XFS directory.
• fix: correct item count and size for empty dirs #574 replaced that with EmptyDirSize (a 512 B apparent-size sentinel) and set directory usage to 0. That removes the wrong number but not the mistake: directories are now charged nothing at all, so on a filesystem that does allocate a block, the directory's own usage is dropped from the total. On ext4 a directory holding two files reports 0 B where du reports 4096 B.

Both directions come from the same place: a directory's usage was assumed, never measured.

Fix

Measure it, once, in the shared platform hook, and fold it into the directory's usage where every analyzer already converges:

• dirOwnUsage(path) reads st_blocks * 512 for the directory itself, per platform (0 where the platform exposes no block information, as before).
• setDirPlatformSpecificAttrs stores it as Dir.OwnUsage.
• dirTotals gains ownUsage, and resolveDirStats adds it to the reported usage in both branches. resolveDirStats remains the single definition of the rule, so the parallel, sequential, top-dir, sqlite and stored analyzers all agree.
• The top-dir fast path (processSubDir), which aggregates separately, does the same.

A directory is now charged exactly its st_blocks: 0 for an XFS inline directory, the real block count elsewhere — the same rule du uses.

Testing

Verified against du on ext4 (-n, --summarize, and --db/sqlite all agree):

case	du	before	after
dir with two files	8192 B	0 B	8192 B
dir + 1-byte files	16384 B	0 B	16384 B
--apparent-size (unchanged)	0 B	0 B	0 B

A multi-block XFS directory reduces to the inline case: st_blocks is 0, so the reported usage is 0, matching du.

New tests:

• TestDirOwnBlocksAreCounted — a directory's reported usage includes its own st_blocks, checked against stat, not an assumed block count.
• TestEmptyDirOwnBlocksAreCounted — the gdu assumes directory takes up 1 block when it doesn't (on XFS) #393 case: a directory with zero blocks reports zero.

Existing tests that encoded the old "directories cost nothing" rule (TestEmptyDirStatsAcrossAnalyzers, TestSqliteAnalyzerIgnoreFileType, TestSqliteAnalyzerTimeFilter, TestTopDirAnalyzerCancellationStopsActiveSubdirectory) now take their expected usage from stat, so they pass on any block size and on any filesystem.

• go test ./... — pass
• golangci-lint run -c .golangci.yml — 0 issues
• gofmt — clean

Fixes #393

— Shivam Maurya

[codecov]

Codecov Report

❌ Patch coverage is 91.66667% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 90.74%. Comparing base (387cf50) to head (356aa73).

Files with missing lines	Patch %	Lines
pkg/analyze/file.go	75.00%	2 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #694   +/-   ##
=======================================
  Coverage   90.73%   90.74%           
=======================================
  Files          76       76           
  Lines        6843     6859   +16     
=======================================
+ Hits         6209     6224   +15     
- Misses        633      634    +1     
  Partials        1        1           

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.