Diff Mode

Diff mode compares two assemblies side by side:

dotsider diff v1.dll v2.dll

Each tab shows both assemblies with differences highlighted. Added items appear in green, removed in red, changed in yellow.

Filters

Press f to cycle through diff filters:

Filter	What it shows
All	Everything from both assemblies
Added	Items only in the right assembly
Removed	Items only in the left assembly
Changed	Items present in both but different

This is useful for reviewing breaking changes between library versions, auditing what changed in a new build, or understanding the impact of a refactoring.

What gets compared

Types are matched by fully qualified name. Methods are matched by declaring type, name, and signature. Assembly references are matched by name.

For matched methods, the diff detects changes across several layers:

• Attributes — visibility, vtable layout, sealed/abstract flags
• Implementation attributes — IL vs native, managed vs unmanaged
• Local variable signatures — local types decoded and compared element-by-element
• Exception regions — try/catch structure, handler offsets, catch types resolved by name
• IL instructions — a normalized walk that resolves metadata token operands (method calls, field accesses, type references, string literals, standalone signatures) to their semantic names before comparing, so metadata table reordering between builds does not produce false positives

The Change column on the Methods tab shows which layers differ: attributes, impl, body, or a combination.

Copy

The Summary tab has selectable info panels and change statistics. Press Tab to cycle focus between them, select text, and press y to copy. iw and yiw work in these panels for quick word-level copying. V selects the entire line and yy copies it.

On the Types, Methods, and References tabs, focus a row and press y to copy it as tab-separated values. A brief flash confirms the yank.

Size diff (Native AOT)

Two mstat-backed inputs — bare .mstat size reports or Native AOT binaries with mstat sidecars — open the size-diff view instead: AOT binaries carry no ECMA-335 metadata, so the managed tabs would show empty tables. What matters for an AOT pair is where the bytes went, and that is what renders.

dotsider diff before.mstat after.mstat
dotsider diff bin/v1/publish/app bin/v2/publish/app
dotsider diff before.mstat after.mstat --json   # headless document instead of the TUI

Inputs	Opens
managed dll ↔ managed dll	the metadata diff above
mstat-backed ↔ mstat-backed	the size-diff view (Summary + Size Map)
AOT binaries without mstat sidecars	the metadata diff, with a hint to publish with IlcGenerateMstatFile
mstat-backed ↔ anything else	an error — the two sides would measure different things

The Size Map tab is a delta treemap: rectangle area is the absolute delta, so the largest regression is the largest rectangle and unchanged mass disappears entirely. Color carries direction — and note it is deliberately the inverse of the managed tabs’ colors, because here red means the binary got bigger:

Color	Meaning
bright red +	added — exists only in the new build
orange Δ	grown
soft green Δ	shrunk
bright green −	removed — exists only in the old build

Direction is never carried by color alone: every rectangle leads with its kind glyph and a signed delta (+12.3 KB). Aggregated entries — overload display collisions, frozen objects grouped by owner — say so with an (n entries) suffix.

Keys: Enter/click drills into a subtree, Esc/right-click goes back up, ←/→ cycle selection, / searches with n/N, and f cycles five direction filters (All, Added, Removed, Grown, Shrunk). w opens the dependency chain that keeps the entry in the binary — added entries answer from the new build’s graph, removed entries from the old build’s, and pressing w again on a changed entry flips sides (the popup header names the side). Namespace, type, assembly, and category tiles roll up the child DGML node names and show representative root-first chains, so you do not have to drill to a method leaf just to ask why a large subtree grew. Entries that truly have no DGML node, such as metadata blobs or resources, say that directly. d on a method disassembles its native body, resolved by the entry’s mangled ILC node name — exact per overload and per instantiation. Repeated presses cycle through every candidate body: an aggregate’s symbols, and for a changed entry both builds — new first, then baseline — so a grown method’s before and after are one key apart. The popup header names the side and which of how many is showing.

The Summary tab prints the same figures the headless size-check report does: totals on every applicable basis, per-kind direction counts, and the top regressions and improvements.

For the CI side of this — budgets, exit codes, and pipeline wiring — see Size Regression.