serialization: fix double-read and type mismatch in exception_ptr load() by arpittkhandelwal · Pull Request #7278 · TheHPXProject/hpx

arpittkhandelwal · 2026-05-19T16:42:14Z

Summary

Two bugs in hpx::serialization::detail::load() for std::exception_ptr in libs/core/serialization/src/exception_ptr.cpp.

Bug 1 — Double-read of archive fields (silent data corruption)

In the load() function, for hpx_exception and std_system_error types, the err_value and err_message fields were read from the archive twice — once via ar & ... and again via ar >> ...:

// Before (buggy):
if (hpx::util::exception_type::hpx_exception == type)
{
    ar & err_value;   // reads from archive (advances cursor)
    ar >> err_value;  // reads AGAIN — cursor double-advanced → garbage!
}
else if (... std_system_error ...)
{
    ar & err_value & err_message;     // reads both
    ar >> err_value >> err_message;   // reads both AGAIN
}

Since save() writes each field once (using ar <<), this caused the deserialization read cursor to advance by 2× what was expected, producing silently wrong err_value and err_message in any distributed HPX application that propagates hpx::exception or std::system_error across localities.

Fix: Remove the redundant ar & ... reads; keep only ar >> ... to match the ar << used in save().

Bug 2 — Type mismatch for `throw_line_` (`long` saved, `int` loaded)

save() declares throw_line_ as long (8 bytes on 64-bit Linux), but load() declared it as int (4 bytes):

// save() — line 45:
long throw_line_ = 0;  // 8 bytes on 64-bit Linux

// load() — line 185 (before fix):
int throw_line_ = 0;   // only 4 bytes read!

On platforms where sizeof(long) != sizeof(int) (e.g. 64-bit Linux/macOS), the serializer writes 8 bytes but the deserializer reads only 4, shifting all subsequent field reads by 4 bytes — a cascading deserialization corruption.

Fix: Change int throw_line_ = 0 → long throw_line_ = 0 in load() to match save().

Changes

libs/core/serialization/src/exception_ptr.cpp

-3 / +1 lines net change

Testing

These bugs are triggered in distributed scenarios where exceptions cross locality boundaries. The fix is mechanical — the load path now mirrors the save path exactly.

codacy-production · 2026-05-19T16:43:12Z

Up to standards ✅

🟢 Issues 0 issues

Results:
0 new issues

View in Codacy

_{NEW Get contextual insights on your PRs based on Codacy's metrics, along with PR and Jira context, without leaving GitHub. Enable AI reviewer}
_{TIP This summary will be updated as you push new changes.}

Copilot

Pull request overview

Fixes two deserialization bugs in hpx::serialization::detail::load() for std::exception_ptr: a double-read of archive fields and a long/int type mismatch for throw_line_. Both caused silent data corruption when exceptions crossed locality boundaries in distributed HPX applications. The load path now mirrors the save path exactly.

Changes:

Removed redundant ar & err_value / ar & err_value & err_message reads that were duplicating the subsequent ar >> ... reads (since operator& on input_archive is equivalent to operator>>).
Changed int throw_line_ = 0 to long throw_line_ = 0 in load() to match the type used in save().

hkaiser

Excellent catch! LGTM, thanks!

hkaiser · 2026-05-19T18:41:48Z

@arpittkhandelwal Could you please construct a test that verifies the fix?

StellarBot · 2026-05-20T02:14:53Z

Performance test report

HPX Performance

Comparison

BENCHMARK	FORK_JOIN_EXECUTOR	PARALLEL_EXECUTOR	SCHEDULER_EXECUTOR
For Each	(=)	(=)	(=)

Info

Property	Before	After
HPX Commit	`0eeca86`	`5decf74`
HPX Datetime	2026-03-09T14:08:29+00:00	2026-05-19T16:42:14+00:00
Hostname	medusa08.rostam.cct.lsu.edu	medusa08.rostam.cct.lsu.edu
Compiler	/opt/apps/llvm/18.1.8/bin/clang++ 18.1.8	/opt/apps/llvm/18.1.8/bin/clang++ 18.1.8
Envfile
Clustername	rostam	rostam
Datetime	2026-03-09T09:15:24.034803-05:00	2026-05-19T21:11:33.411763-05:00

Comparison

BENCHMARK	NO-EXECUTOR
Future Overhead - Create Thread Hierarchical - Latch	++

Info

Property	Before	After
HPX Commit	`0eeca86`	`5decf74`
HPX Datetime	2026-03-09T14:08:29+00:00	2026-05-19T16:42:14+00:00
Hostname	medusa08.rostam.cct.lsu.edu	medusa08.rostam.cct.lsu.edu
Compiler	/opt/apps/llvm/18.1.8/bin/clang++ 18.1.8	/opt/apps/llvm/18.1.8/bin/clang++ 18.1.8
Envfile
Clustername	rostam	rostam
Datetime	2026-03-09T09:17:15.638328-05:00	2026-05-19T21:13:41.524568-05:00

Comparison

BENCHMARK	FORK_JOIN_EXECUTOR_DEFAULT_FORK_JOIN_POLICY_ALLOCATOR	PARALLEL_EXECUTOR_DEFAULT_PARALLEL_POLICY_ALLOCATOR	SCHEDULER_EXECUTOR_DEFAULT_SCHEDULER_EXECUTOR_ALLOCATOR
Stream Benchmark - Add	(=)	(=)	--
Stream Benchmark - Scale	(=)	--	--
Stream Benchmark - Triad	(=)	-	--
Stream Benchmark - Copy	(=)	+++	++

Info

Property	Before	After
HPX Commit	`ba89f5d`	`5decf74`
HPX Datetime	2026-03-09T18:50:37+00:00	2026-05-19T16:42:14+00:00
Hostname	medusa08.rostam.cct.lsu.edu	medusa08.rostam.cct.lsu.edu
Compiler	/opt/apps/llvm/18.1.8/bin/clang++ 18.1.8	/opt/apps/llvm/18.1.8/bin/clang++ 18.1.8
Envfile
Clustername	rostam	rostam
Datetime	2026-03-09T17:49:10.837937-05:00	2026-05-19T21:14:26.875804-05:00

Explanation of Symbols

Symbol	MEANING
=	No performance change (confidence interval within ±1%)
(=)	Probably no performance change (confidence interval within ±2%)
(+)/(-)	Very small performance improvement/degradation (≤1%)
+/-	Small performance improvement/degradation (≤5%)
++/--	Large performance improvement/degradation (≤10%)
+++/---	Very large performance improvement/degradation (>10%)
?	Probably no change, but quite large uncertainty (confidence interval with ±5%)
??	Unclear result, very large uncertainty (±10%)
???	Something unexpected…

Two bugs in hpx::serialization::detail::load() for std::exception_ptr: 1. Double-read of archive fields (Bug TheHPXProject#1 - data corruption): In the load() function, for hpx_exception and std_system_error types, the err_value and err_message fields were read from the archive twice: once via 'ar & ...' and again via 'ar >> ...'. Since save() writes each field only once, this caused the read cursor to advance by 2x, silently producing garbled error codes and messages in any distributed HPX application that propagates these exception types across localities. Fix: Remove the redundant 'ar & ...' reads; keep only 'ar >> ...' which matches the 'ar << ...' used in save(). 2. Type mismatch for throw_line_ (Bug TheHPXProject#3 - platform-specific corruption): save() declares throw_line_ as 'long' (8 bytes on 64-bit Linux), but load() declared it as 'int' (4 bytes). This caused the serializer to write 8 bytes and the deserializer to read only 4, shifting all subsequent field reads by 4 bytes on affected platforms. Fix: Change 'int throw_line_ = 0' to 'long throw_line_ = 0' in load() to match the type used in save(). Additionally, added a regression test to verify that serialization of hpx::exception, std::system_error, std::runtime_error, and sequential round-tripping behaves correctly. Signed-off-by: arpittkhandelwal <arpitkhandelwal810@gmail.com>

arpittkhandelwal · 2026-05-20T03:33:11Z

@arpittkhandelwal Could you please construct a test that verifies the fix?

Yes I have constructed and added a comprehensive regression unit test under libs/core/serialization/tests/unit/serialization_exception_ptr.cpp to verify these fixes.

StellarBot · 2026-05-20T03:45:25Z

Performance test report

HPX Performance

Comparison

BENCHMARK	FORK_JOIN_EXECUTOR	PARALLEL_EXECUTOR	SCHEDULER_EXECUTOR
For Each	(=)	(=)	(=)

Info

Property	Before	After
HPX Datetime	2026-03-09T14:08:29+00:00	2026-05-20T03:32:29+00:00
HPX Commit	`0eeca86`	`094caac`
Datetime	2026-03-09T09:15:24.034803-05:00	2026-05-19T22:41:57.886090-05:00
Compiler	/opt/apps/llvm/18.1.8/bin/clang++ 18.1.8	/opt/apps/llvm/18.1.8/bin/clang++ 18.1.8
Envfile
Clustername	rostam	rostam
Hostname	medusa08.rostam.cct.lsu.edu	medusa08.rostam.cct.lsu.edu

Comparison

BENCHMARK	NO-EXECUTOR
Future Overhead - Create Thread Hierarchical - Latch	++

Info

Property	Before	After
HPX Datetime	2026-03-09T14:08:29+00:00	2026-05-20T03:32:29+00:00
HPX Commit	`0eeca86`	`094caac`
Datetime	2026-03-09T09:17:15.638328-05:00	2026-05-19T22:44:07.143095-05:00
Compiler	/opt/apps/llvm/18.1.8/bin/clang++ 18.1.8	/opt/apps/llvm/18.1.8/bin/clang++ 18.1.8
Envfile
Clustername	rostam	rostam
Hostname	medusa08.rostam.cct.lsu.edu	medusa08.rostam.cct.lsu.edu

Comparison

BENCHMARK	FORK_JOIN_EXECUTOR_DEFAULT_FORK_JOIN_POLICY_ALLOCATOR	PARALLEL_EXECUTOR_DEFAULT_PARALLEL_POLICY_ALLOCATOR	SCHEDULER_EXECUTOR_DEFAULT_SCHEDULER_EXECUTOR_ALLOCATOR
Stream Benchmark - Add	(=)	=	--
Stream Benchmark - Scale	(=)	--	--
Stream Benchmark - Triad	(=)	-	--
Stream Benchmark - Copy	(=)	+++	++

Info

Property	Before	After
HPX Datetime	2026-03-09T18:50:37+00:00	2026-05-20T03:32:29+00:00
HPX Commit	`ba89f5d`	`094caac`
Datetime	2026-03-09T17:49:10.837937-05:00	2026-05-19T22:44:53.173799-05:00
Compiler	/opt/apps/llvm/18.1.8/bin/clang++ 18.1.8	/opt/apps/llvm/18.1.8/bin/clang++ 18.1.8
Envfile
Clustername	rostam	rostam
Hostname	medusa08.rostam.cct.lsu.edu	medusa08.rostam.cct.lsu.edu

Explanation of Symbols

Symbol	MEANING
=	No performance change (confidence interval within ±1%)
(=)	Probably no performance change (confidence interval within ±2%)
(+)/(-)	Very small performance improvement/degradation (≤1%)
+/-	Small performance improvement/degradation (≤5%)
++/--	Large performance improvement/degradation (≤10%)
+++/---	Very large performance improvement/degradation (>10%)
?	Probably no change, but quite large uncertainty (confidence interval with ±5%)
??	Unclear result, very large uncertainty (±10%)
???	Something unexpected…

hkaiser · 2026-05-20T13:01:20Z

+{
+    // Construct an hpx::exception whose internal throw_line will be stored
+    // as a long; the exact value is set by the HPX_THROW_EXCEPTION macro but
+    // we can still verify the type and error code survive.


You could still use hpx::detail::throw_exception directly and pass in a large line number for testing, see:

hpx/libs/core/errors/include/hpx/errors/macros.hpp

Lines 105 to 106 in ed201ce

hpx::detail::throw_exception( \

errcode, hpx::util::format(__VA_ARGS__), f, __FILE__, __LINE__) /**/

hkaiser · 2026-05-20T14:13:39Z

The new test looks good, it however doesn't compile :/

Also, your LLM use caught up with you again:

/libs/core/serialization/tests/unit/serialization_exception_ptr.cpp:
    Non-ASCII: line 10

arpittkhandelwal requested a review from hkaiser as a code owner May 19, 2026 16:42

Copilot AI review requested due to automatic review settings May 19, 2026 16:42

Copilot started reviewing on behalf of arpittkhandelwal May 19, 2026 16:42 View session

Copilot AI reviewed May 19, 2026

View reviewed changes

hkaiser previously approved these changes May 19, 2026

View reviewed changes

hkaiser added type: enhancement type: compatibility issue category: serialization category: exception-handling labels May 19, 2026

hkaiser added this to the 2.0.0 milestone May 19, 2026

arpittkhandelwal dismissed hkaiser’s stale review via 6993160 May 20, 2026 03:31

arpittkhandelwal force-pushed the fix/exception-ptr-serialization-double-read branch from 7703618 to 6993160 Compare May 20, 2026 03:31

arpittkhandelwal force-pushed the fix/exception-ptr-serialization-double-read branch from 6993160 to 0be64ba Compare May 20, 2026 03:32

hkaiser reviewed May 20, 2026

View reviewed changes

	hpx::detail::throw_exception( \
	errcode, hpx::util::format(__VA_ARGS__), f, __FILE__, __LINE__) /**/

Uh oh!

Conversation

arpittkhandelwal commented May 19, 2026

Summary

Bug 1 — Double-read of archive fields (silent data corruption)

Bug 2 — Type mismatch for throw_line_ (long saved, int loaded)

Changes

Testing

Uh oh!

codacy-production Bot commented May 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Up to standards ✅

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

hkaiser left a comment

Choose a reason for hiding this comment

Uh oh!

hkaiser commented May 19, 2026

Uh oh!

StellarBot commented May 20, 2026

HPX Performance

Comparison

Info

Comparison

Info

Comparison

Info

Explanation of Symbols

Uh oh!

arpittkhandelwal commented May 20, 2026

Uh oh!

StellarBot commented May 20, 2026

HPX Performance

Comparison

Info

Comparison

Info

Comparison

Info

Explanation of Symbols

Uh oh!

hkaiser May 20, 2026

Choose a reason for hiding this comment

Uh oh!

hkaiser commented May 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Bug 2 — Type mismatch for `throw_line_` (`long` saved, `int` loaded)

codacy-production Bot commented May 19, 2026 •

edited

Loading

hkaiser commented May 20, 2026 •

edited

Loading