Fix debug info for wide pointers #27969

jabraham17 · 2025-10-27T23:50:43Z

Fixes the debug info for wide pointers, by correctly creating debug info for localeIDs.

Having create debug info for wide pointers allows the pretty printer to properly resolve wide pointers, which this PR also implements. The pretty printer uses a new runtime function to submit GETs to resolve wide pointers. This enables pretty-printing domains and arrays with --no-local or CHPL_COMM!=none. However, future work is still needed to handle true distributed arrays (arrays with data on more than one locale)

start_test test/llvm/debugInfo/
start_test test/llvm/debugInfo/ with --no-local
start_test test/llvm/debugInfo/dwarfdump/ with COMM=gasnet
paratest with/without gasnet

Resolves #27778

Signed-off-by: Jade Abraham <[email protected]>

DanilaFe · 2025-11-06T17:36:13Z

compiler/llvm/llvmDebug.cpp

+  } else if (type == dtLocaleID) {
+    // handle locale types
+    llvm::SmallVector<llvm::Metadata *, 1> EltTys;
+    llvm::Type* nodeTy = dtInt[INT_SIZE_32]->getLLVMType();


Is it well-established that locales are backed by int32? is there somewhere in the compiler you can reference to be sure that it's dtInt[INT_SIZE_32]

That would be ideal. info->lvt->getType("c_nodeid_t") should do the trick for this case

DanilaFe · 2025-11-06T17:40:31Z

compiler/llvm/llvmDebug.cpp

+        layout.getTypeSizeInBits(nodeTy),
+        8*layout.getABITypeAlign(nodeTy).value(),
+        layout.getTypeSizeInBits(nodeTy), /* offset, assume after node */
+        llvm::DINode::FlagZero,


I see we do similar things below for other types, but how do we know this is the alignment of the locale type? Can we be sure?

We don't know for certain, to know for certain we would need to compute the offset from the GEP to access subloc. But technically Chapel doesn't know that info (its an opaque record), so there is no GEP (because there is no member access).

But its a pretty good assumption to make. In a struct with 2 primitives of the same type the offset of the second var will just be the size of the primitive type

DanilaFe · 2025-11-06T17:40:48Z

compiler/llvm/llvmDebug.cpp

+      8*layout.getABITypeAlign(ty).value(), /* AlignInBits */
+      llvm::DINode::FlagZero, /* Flags */
+      nullptr,
+      dibuilder->getOrCreateArray(EltTys) /* Elements */


Which of these lines are you suggesting std::move should go on?

Sorry, EltTys

The function takes an llvm::ArrayRef, which is a constant reference. I don't think the move is needed

runtime/etc/debug/chpl_lldb_pretty_print.py

test/llvm/debugInfo/dwarfdump/arrays.comm-none.good

DanilaFe · 2025-11-06T18:01:10Z

test/llvm/debugInfo/dwarfdump/domains.good

 DW_TAG_structure_type
              DW_AT_name	("DefaultDist")
-              DW_AT_byte_size	(0x18)
+              DW_AT_byte_size	(0x20)


Curious why this number might have changed?

Because wide pointers affect the size/alignment

DanilaFe · 2025-11-06T18:03:37Z

test/llvm/debugInfo/gdb/basicTypes.good

@@ -1,8 +1,8 @@
 Reading symbols from basicTypes...
-Breakpoint 1 at 0xXXXX: file debugger.c, line 28.
+Breakpoint 1 at 0xXXXX: file debugger.c, line 31.


should there be a prediff for line numbers?

These line numbers are the lines of the file basicTypes.chpl, and part of what this test is locking in is the line numbers. So no

DanilaFe · 2025-11-06T18:05:26Z

test/llvm/debugInfo/gdb/SKIPIF

+if [[ "$CHPL_COMM" != "none" ]]; then
+  echo "True"
+  exit 0
+fi
+


Because --gdb and --lldb work differently with COMM!=none

DanilaFe · 2025-11-06T18:09:37Z

test/llvm/debugInfo/lldb/basicTypes.good

-    dom = 0xXXXX
-    data = 0xXXXX


This is because you removed the synthetic fields (you mentioned "array gets confused" in the comments)?

Signed-off-by: Jade Abraham <[email protected]>

jabraham17 · 2025-11-07T01:01:24Z

Running testing after apply some reviewer feedback shows test/llvm/debugInfo/lldb/basicTypes.chpl is now failing pretty badly for --no-local. Not sure when that slipped in. But I need to fix that before merging

jabraham17 marked this pull request as draft October 27, 2025 23:51

jabraham17 force-pushed the debug-locale_t branch 3 times, most recently from 11a41ca to ce541b4 Compare October 30, 2025 18:40

jabraham17 added 17 commits November 3, 2025 11:30

add debug info for locales and wide pointers

6047347

Signed-off-by: Jade Abraham <[email protected]>

expand classes testcase

c21b741

Signed-off-by: Jade Abraham <[email protected]>

remove suppressif

7e08aa2

Signed-off-by: Jade Abraham <[email protected]>

add dummy files for wide pointers

cc528af

Signed-off-by: Jade Abraham <[email protected]>

add warning

26a09da

Signed-off-by: Jade Abraham <[email protected]>

add array wide test cases

4098112

Signed-off-by: Jade Abraham <[email protected]>

prevent crashes due to wide pointers

15d9d81

Signed-off-by: Jade Abraham <[email protected]>

fix array type lookup

a355fd1

Signed-off-by: Jade Abraham <[email protected]>

get arrays with wide pointers working properly

629f993

Signed-off-by: Jade Abraham <[email protected]>

remove suppressifs

4dd0b5d

Signed-off-by: Jade Abraham <[email protected]>

init arrays

ec671cf

Signed-off-by: Jade Abraham <[email protected]>

handle wide pointers for domains

7609aab

Signed-off-by: Jade Abraham <[email protected]>

handle owned/shared classes

67698b9

Signed-off-by: Jade Abraham <[email protected]>

handle owned/shared classes

120f738

Signed-off-by: Jade Abraham <[email protected]>

remove data and dom

815e7c0

Signed-off-by: Jade Abraham <[email protected]>

add initial basicTypes.no-local.good

c77740c

Signed-off-by: Jade Abraham <[email protected]>

update good file with wide pointers

737ac6d

Signed-off-by: Jade Abraham <[email protected]>

jabraham17 force-pushed the debug-locale_t branch from f0033e4 to 737ac6d Compare November 3, 2025 19:30

jabraham17 added 2 commits November 3, 2025 15:13

avoid chapel-lang#28008

45cf337

Signed-off-by: Jade Abraham <[email protected]>

update good files

09564fd

Signed-off-by: Jade Abraham <[email protected]>

jabraham17 marked this pull request as ready for review November 3, 2025 23:19

jabraham17 added 3 commits November 3, 2025 18:28

fix gdb good files

411db6b

Signed-off-by: Jade Abraham <[email protected]>

fix dwarf target detection

700960a

Signed-off-by: Jade Abraham <[email protected]>

prediff comm specific paths

3f7a93f

Signed-off-by: Jade Abraham <[email protected]>

jabraham17 mentioned this pull request Nov 4, 2025

Fix dSYM generation when CHPL_LAUNCHER!=none #28009

Open

jabraham17 requested a review from DanilaFe November 4, 2025 00:34

DanilaFe approved these changes Nov 6, 2025

View reviewed changes

jabraham17 added 5 commits November 6, 2025 13:21

use c_nodeid_t

1205923

Signed-off-by: Jade Abraham <[email protected]>

use c_sublocid_t

bdd576d

Signed-off-by: Jade Abraham <[email protected]>

use ptr instead of wideptr

2dc69a8

Signed-off-by: Jade Abraham <[email protected]>

fix error handling

9562aad

Signed-off-by: Jade Abraham <[email protected]>

use get()

de42928

Signed-off-by: Jade Abraham <[email protected]>

Fix debug info for wide pointers #27969

Are you sure you want to change the base?

Fix debug info for wide pointers #27969

Uh oh!

Conversation

jabraham17 commented Oct 27, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jabraham17 commented Nov 7, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

jabraham17 commented Oct 27, 2025 •

edited

Loading