Split gc.cpp to multiple files based on the functionality - part 1 by janvorli · Pull Request #125703 · dotnet/runtime

janvorli · 2026-03-18T00:54:12Z

This PR splits the monolithic gc.cpp (~57,000 lines) into 19 smaller files organized by functionality, reducing gc.cpp to ~8,300 lines of shared infrastructure. The new files are #included at the end of gc.cpp, preserving the
existing single-compilation-unit build model for this first step - no CMakeLists.txt changes were made.

New files

File	Lines	Scope
`allocation.cpp`	4,337	Allocation helpers, `allocate_small/large`, SOH/LOH/POH allocation
`background.cpp`	3,964	Background GC (concurrent mark, sweep, write watch support)
`card_table.cpp`	1,810	Card/brick table management, card bundling
`collect.cpp`	1,511	Collection entry points, `garbage_collect`, `do_post_gc`
`diagnostics.cpp`	1,615	Heap verification, `verify_heap`, ETW/diag walking
`dynamic_heap_count.cpp`	1,280	Dynamic heap count (server GC heap scaling)
`dynamic_tuning.cpp`	2,441	GC tuning heuristics, generation budgets, smoothing
`finalization.cpp`	520	Finalization queue management
`init.cpp`	1,350	GC initialization (`gc_heap::initialize_gc`)
`interface.cpp`	2,261	`GCHeap` interface implementation (public API surface)
`mark_phase.cpp`	3,728	Mark phase, mark stack, pinned object handling
`memory.cpp`	420	Virtual memory commit/decommit, address space management
`no_gc.cpp`	827	No-GC region support
`plan_phase.cpp`	7,374	Plan phase, plug/gap processing, generation planning
`region_allocator.cpp`	421	Region allocator (regions mode)
`region_free_list.cpp`	420	Region free list management
`regions_segments.cpp`	2,052	Segment/region lifecycle, segment mapping table
`relocate_compact.cpp`	1,977	Relocate and compact phases
`sweep.cpp`	527	Sweep phase, free list building

Design decisions

Single compilation unit preserved: All new files are #included from gc.cpp, so gcwks.cpp and gcsvr.cpp continue to compile everything as one TU. No build system changes required.
#ifdef guards maintained: Each moved method retains its original preprocessor guards (BACKGROUND_GC, USE_REGIONS, MULTIPLE_HEAPS, etc.).
- Remaining in gc.cpp: Shared globals, macros, general-purpose helpers, and methods that don't clearly belong to one functional category stay in gc.cpp.

What's NOT changed

No functional/behavioral changes — this is a pure code reorganization
No changes to headers (gcpriv.h, gc.h, etc.)
No changes to the build system (CMakeLists.txt)
All existing #ifdef nesting is preserved

These changes were made by copilot cli with my supervision and reviewing. The next step will be to actually compile each of the files separately.

dotnet-policy-service · 2026-03-18T00:55:09Z

Tagging subscribers to this area: @agocke, @dotnet/gc
See info in area-owners.md if you want to be subscribed.

Copilot

Pull request overview

This pull request appears to refactor CoreCLR GC implementation by moving substantial logic into new .cpp compilation units (which are then included from gc.cpp), separating concerns like sweeping, regions, no-GC mode, memory commit/decommit accounting, collection, initialization, and finalization.

Changes:

Adds new GC implementation files for sweeping, regions (allocator/free list), no-GC region behavior, memory commit/decommit, collection, initialization, and finalization.
Updates GC internals by removing an unused VERIFY_HEAP method declaration from gcpriv.h.
(Implied by file structure) Consolidates GC implementation via #include-based composition of these new .cpp units.

Reviewed changes

Copilot reviewed 15 out of 21 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
src/coreclr/gc/sweep.cpp	Adds sweep/free-list building and related helpers (incl. FEATURE_BASICFREEZE RO sweep).
src/coreclr/gc/region_free_list.cpp	Adds region free-list management for `USE_REGIONS`.
src/coreclr/gc/region_allocator.cpp	Adds region allocation/deallocation logic for `USE_REGIONS`.
src/coreclr/gc/no_gc.cpp	Adds no-GC region support logic, including callback scheduling and region-extension helpers.
src/coreclr/gc/memory.cpp	Adds commit/decommit accounting and region decommit logic.
src/coreclr/gc/init.cpp	Adds GC initialization logic (including region range reservation/init under `USE_REGIONS`).
src/coreclr/gc/finalization.cpp	Adds finalization queue implementation and finalizer work scheduling.
src/coreclr/gc/collect.cpp	Adds GC collection logic (including region ephemeral/gc range computation under `USE_REGIONS`).
src/coreclr/gc/gcpriv.h	Removes `set_batch_mark_array_bits` declaration under `VERIFY_HEAP`.

You can also share your feedback on Copilot code review. Take the survey.

src/coreclr/gc/sweep.cpp

+#ifndef USE_REGIONS
+    if ((settings.condemned_generation == max_generation) && ro_segments_in_range)
+    {
+        heap_segment* seg = generation_start_segment (generation_of (max_generation));;


src/coreclr/gc/no_gc.cpp

+    dprintf(1, ("[no_gc_callback] calling enable_no_gc_callback with callback_threshold = %llu\n", callback_threshold));
+    enable_no_gc_region_callback_status status = enable_no_gc_region_callback_status::succeed;
+    suspend_EE();
+    {
+        if (!current_no_gc_region_info.started)
+        {
+            status = enable_no_gc_region_callback_status::not_started;
+        }
+        else if (current_no_gc_region_info.callback != nullptr)
+        {
+            status = enable_no_gc_region_callback_status::already_registered;
+        }


src/coreclr/gc/no_gc.cpp

+                        status = insufficient_budget;
+                    }
+                    if (dd_new_allocation (hp->dynamic_data_of (loh_generation)) <= (ptrdiff_t)loh_withheld_budget)
+                    {
+                        dprintf(1, ("[no_gc_callback] failed because of running out of loh budget= %llu\n", loh_withheld_budget));
+                        status = insufficient_budget;


src/coreclr/gc/region_allocator.cpp

+    *start = alloc;
+    *end = alloc + alloc_size;
+    ret = (alloc != NULL);
+
+    gc_etw_segment_type segment_type;
+
+    if (gen_num == loh_generation)
+    {
+        segment_type = gc_etw_segment_large_object_heap;
+    }
+    else if (gen_num == poh_generation)
+    {
+        segment_type = gc_etw_segment_pinned_object_heap;
+    }
+    else
+    {
+        segment_type = gc_etw_segment_small_object_heap;
+    }
+
+    FIRE_EVENT(GCCreateSegment_V1, (alloc + sizeof (aligned_plug_and_gap)),
+                                  size - sizeof (aligned_plug_and_gap),
+                                  segment_type);
+


src/coreclr/gc/region_allocator.cpp

+        {
+            uint32_t next_size = get_num_units (next_val);
+            free_block_size += next_size;
+            region_end += next_size;


src/coreclr/gc/region_allocator.cpp

+                    uint32_t* busy_block;
+                    uint32_t* free_block;
+                    if (direction == 1)
+                    {
+                        busy_block = current_index;
+                        free_block = current_index + num_units;
+                    }
+                    else
+                    {
+                        busy_block = current_index - num_units;
+                        free_block = current_index - current_num_units;
+                    }
+


MichalStrehovsky · 2026-03-18T01:40:56Z

If there is interest in keeping git blame working, it might be possible to commit this in a way that preserves it: https://devblogs.microsoft.com/oldnewthing/20190916-00/?p=102892

A repo admin might be necessary to actually merge that because a squash would likely undo it.

janvorli · 2026-03-18T01:59:48Z

If there is interest in keeping git blame working, it might be possible to commit this in a way that preserves it: https://devblogs.microsoft.com/oldnewthing/20190916-00/?p=102892

That sounds very interesting. I'll give that a try.

jkoritzinsky · 2026-03-18T02:01:24Z

If necessary for perf reasons (ie lost optimizations that LTCG doesn’t figure out when you try to compile separately in the next PR), you can use the UNITY_BUILD feature in CMake to force all of the files to be included in one C++ file in Release builds to maintain the current experience while providing a better dev story.

janvorli · 2026-03-18T12:36:11Z

@MichalStrehovsky it seems that git blame still works without any special handling. You just need to pass it the -C -C option. Copilot says github UI doesn't use that option though, so it would only work from the command line. Not sure if the way suggested in the article you've shared would work in github too. But i'll give it a try anyways. Btw, the blame in the github UI never worked for the gc.cpp, as it always times out because the file is so huge. So I wonder if it would have any benefit even if I did that based on the article.

janvorli · 2026-03-18T12:43:55Z

Hmm, I take it back, the git -C -C doesn't work.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split the monolithic gc.cpp (~57,000 lines) into 19 smaller files organized by functionality, reducing gc.cpp to ~8,300 lines of core infrastructure. New files are #included at the end of gc.cpp, preserving the single-compilation-unit build model. Uses rename-per-branch technique to preserve git blame history across the split. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

The gc.cpp split uses rename-per-branch to preserve blame history. Adding the trim commits to .git-blame-ignore-revs lets git blame (and GitHub's web UI) see through them to the original gc.cpp authors. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

Pull request overview

This PR appears to split previously monolithic GC implementation code into multiple focused .cpp files under src/coreclr/gc/ (e.g., sweep/regions/no-GC/memory/init/finalization/collect) and updates .git-blame-ignore-revs to keep git blame attribution useful after the mechanical refactor.

Changes:

Extract GC implementation areas into new compilation-unit fragments (included from gc.cpp), such as sweeping, region allocation/free-list management, no-GC region logic, memory commit/decommit accounting, GC init, finalization, and collection logic.
Add region allocator/free-list implementations behind USE_REGIONS.
Add blame-ignore entries for the mechanical “split gc.cpp” trimming commits.

Reviewed changes

Copilot reviewed 15 out of 21 changed files in this pull request and generated 8 comments.

Show a summary per file

File	Description
src/coreclr/gc/sweep.cpp	New split-out sweeping/free-list threading logic (plus RO segment sweeping under BASICFREEZE).
src/coreclr/gc/region_free_list.cpp	New region free-list bookkeeping, ordering, and sorting utilities (USE_REGIONS).
src/coreclr/gc/region_allocator.cpp	New region allocator implementation (USE_REGIONS). Contains correctness issues noted in PR comments.
src/coreclr/gc/no_gc.cpp	New split-out no-GC region and callback logic.
src/coreclr/gc/memory.cpp	New split-out commit/decommit accounting and decommit stepping logic (incl. regions).
src/coreclr/gc/init.cpp	New split-out GC initialization logic (incl. regions initial reservation).
src/coreclr/gc/finalization.cpp	New split-out finalization queue logic and finalizer work scheduling.
src/coreclr/gc/collect.cpp	New split-out GC collection driver logic and region ephemeral-range computation.
.git-blame-ignore-revs	Adds mechanical split/trim commits to blame-ignore list to preserve attribution.

You can also share your feedback on Copilot code review. Take the survey.

src/coreclr/gc/region_allocator.cpp

+    alloc = allocate (num_units, direction, fn);
+    *start = alloc;
+    *end = alloc + alloc_size;
+    ret = (alloc != NULL);


src/coreclr/gc/region_allocator.cpp

+    FIRE_EVENT(GCCreateSegment_V1, (alloc + sizeof (aligned_plug_and_gap)),
+                                  size - sizeof (aligned_plug_and_gap),
+                                  segment_type);


src/coreclr/gc/region_allocator.cpp

+        {
+            uint32_t next_size = get_num_units (next_val);
+            free_block_size += next_size;
+            region_end += next_size;


src/coreclr/gc/region_allocator.cpp

+    uint32_t current_val = *current_index;
+    assert (!is_unit_memory_free (current_val));
+
+    dprintf (REGIONS_LOG, ("----DEL %d (%u units)-----", (*current_index - *region_map_left_start), current_val));


src/coreclr/gc/region_allocator.cpp

+                    uint32_t* busy_block;
+                    uint32_t* free_block;
+                    if (direction == 1)
+                    {
+                        busy_block = current_index;
+                        free_block = current_index + num_units;
+                    }
+                    else


src/coreclr/gc/no_gc.cpp

+                        dprintf(1, ("[no_gc_callback] failed because of running out of soh budget= %llu\n", soh_withheld_budget));
+                        status = insufficient_budget;
+                    }
+                    if (dd_new_allocation (hp->dynamic_data_of (loh_generation)) <= (ptrdiff_t)loh_withheld_budget)
+                    {
+                        dprintf(1, ("[no_gc_callback] failed because of running out of loh budget= %llu\n", loh_withheld_budget));


src/coreclr/gc/no_gc.cpp

+                        dprintf(1, ("[no_gc_callback] failed because of running out of soh budget= %llu\n", soh_withheld_budget));
+                        status = insufficient_budget;
+                    }
+                    if (dd_new_allocation (hp->dynamic_data_of (loh_generation)) <= (ptrdiff_t)loh_withheld_budget)
+                    {
+                        dprintf(1, ("[no_gc_callback] failed because of running out of loh budget= %llu\n", loh_withheld_budget));


src/coreclr/gc/sweep.cpp

+#ifndef USE_REGIONS
+    if ((settings.condemned_generation == max_generation) && ro_segments_in_range)
+    {
+        heap_segment* seg = generation_start_segment (generation_of (max_generation));;


janvorli · 2026-03-18T13:52:24Z

Copilot cli has reworked the PR based on the article mentioned. It also added a .git-blame-ignore-revs file that enables github blame to work properly by ignoring the changes in this PR in the blame.

To make local git blame work the best, git config diff.algorithm histogram can be used. The default diff is not quite good in properly detecting all moves.

janvorli · 2026-03-18T13:59:49Z

It requires git version >= 2.40 (older git doesn't seem to use the algorithm)

janvorli · 2026-03-18T14:05:29Z

And this PR needs to be merged without squashing. Otherwise the tracking would get lost

janvorli added this to the 11.0.0 milestone Mar 18, 2026

janvorli self-assigned this Mar 18, 2026

Copilot AI review requested due to automatic review settings March 18, 2026 00:54

janvorli added the area-GC-coreclr label Mar 18, 2026

Copilot started reviewing on behalf of janvorli March 18, 2026 00:56 View session

Copilot AI reviewed Mar 18, 2026

View reviewed changes

This was referenced Mar 18, 2026

Cannot find 'arm64-v8a' device dotnet/dnceng#2284

Open

XHarness package install failure on iOS due to devicectl NSPOSIXErrorDomain error 49 #123796

Open

MsQuic fails with QUIC_STATUS_OUT_OF_MEMORY on AzureLinux #123216

Open

janvorli and others added 15 commits March 18, 2026 13:56

Split gc.cpp: rename to region_allocator.cpp

5c2fd0d

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Prepare gc.cpp for split (temporary rename to gc_full.cpp)

2fdb93f

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: trim region_allocator.cpp

890969c

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: rename to region_free_list.cpp

8f6788f

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: rename to finalization.cpp

b45869c

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: trim region_free_list.cpp

1da09f4

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: rename to interface.cpp

4f44ad2

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: trim interface.cpp

158e4b3

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: trim finalization.cpp

ca476dd

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: rename to allocation.cpp

925bcf7

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: trim allocation.cpp

33f469d

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: rename to mark_phase.cpp

abda75b

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: rename to plan_phase.cpp

bcf6c72

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: trim mark_phase.cpp

95fe951

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: rename to relocate_compact.cpp

00356b9

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

janvorli and others added 20 commits March 18, 2026 13:56

Split gc.cpp: trim regions_segments.cpp

1712182

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: rename to memory.cpp

deea9a0

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: trim memory.cpp

82e97a7

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: trim card_table.cpp

59ed3a8

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: rename to diagnostics.cpp

57f2d45

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: trim diagnostics.cpp

f250cb6

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: rename to dynamic_tuning.cpp

6f3d262

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: rename to no_gc.cpp

6333022

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: trim dynamic_tuning.cpp

b092b7b

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: rename to dynamic_heap_count.cpp

cef87be

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: trim no_gc.cpp

dcbaeec

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: rename to init.cpp

81bc353

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: trim init.cpp

43751a9

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: trim dynamic_heap_count.cpp

959aac5

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: rename to collect.cpp

38eba9f

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: trim collect.cpp

958ef80

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: rename gc_full.cpp back to gc.cpp

c817b0d

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Split gc.cpp: trim to core infrastructure with #includes

3ffbdd2

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot AI review requested due to automatic review settings March 18, 2026 13:41

janvorli force-pushed the gc-split branch from 80889f7 to 80e5440 Compare March 18, 2026 13:41

Copilot started reviewing on behalf of janvorli March 18, 2026 13:42 View session

Copilot AI reviewed Mar 18, 2026

View reviewed changes

jkoritzinsky added the NO-SQUASH The PR should not be squashed label Mar 18, 2026

janvorli requested review from VSadov and jkotas March 18, 2026 15:44

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Split gc.cpp to multiple files based on the functionality - part 1#125703

Split gc.cpp to multiple files based on the functionality - part 1#125703
janvorli wants to merge 43 commits intodotnet:mainfrom
janvorli:gc-split

janvorli commented Mar 18, 2026

Uh oh!

dotnet-policy-service bot commented Mar 18, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

MichalStrehovsky commented Mar 18, 2026

Uh oh!

janvorli commented Mar 18, 2026

Uh oh!

jkoritzinsky commented Mar 18, 2026

Uh oh!

janvorli commented Mar 18, 2026

Uh oh!

janvorli commented Mar 18, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

janvorli commented Mar 18, 2026

Uh oh!

janvorli commented Mar 18, 2026

Uh oh!

janvorli commented Mar 18, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

janvorli commented Mar 18, 2026

Uh oh!

dotnet-policy-service bot commented Mar 18, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

MichalStrehovsky commented Mar 18, 2026

Uh oh!

janvorli commented Mar 18, 2026

Uh oh!

jkoritzinsky commented Mar 18, 2026

Uh oh!

janvorli commented Mar 18, 2026

Uh oh!

janvorli commented Mar 18, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

janvorli commented Mar 18, 2026

Uh oh!

janvorli commented Mar 18, 2026

Uh oh!

janvorli commented Mar 18, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants