Creating multiple targets leads to a b2 crash in make1c()

[pkl97]

Environment and version details

• Operating System+version: Red Hat Enterprise Linux 10
• Compiler+version: clang 21
• Shell: Bash
• B2 Version: 5.3.2

Running specific gRPC generation rules in our make system sometimes triggers the following crash in b2:

Core was generated by `/home/user/devel/bin/b2 -f Jamfile -a -j8'.
Program terminated with signal SIGSEGV, Segmentation fault.
#0  0x000055d8ac530970 in make1c (pState=0x55d8b427f110) at make1.cpp:710

(gdb) bt
#0  0x000055d8ac530970 in make1c (pState=0x55d8b427f110) at make1.cpp:710
#1  0x000055d8ac52f8a4 in make1 (targets=0x55d8b4272f10) at make1.cpp:247
#2  0x000055d8ac52d70e in make (targets=0x55d8b4272f10, anyhow=true) at make.cpp:159
#3  0x000055d8ac514391 in guarded_main (argc=5, argv=0x7ffdf5970a58) at jam.cpp:633
#4  0x000055d8ac514863 in main (argc=5, argv=0x7ffdf5970a58) at jam.cpp:694

This sample project shows the issue:

MultipleTargetCrashes.zip

The reproduction is simple:

for i in {1..1000}; do echo $i; b2 -f Jamfile -a -j8; done

One however needs a machine with sufficient parallelism to trigger the crashes. During 1000 runs I obtained 13 crashes.

The reason for the crash is a null pointer dereference in function make1c() of file make1.cpp.

This simple patch fixes the issue:

diff --git a/src/engine/make1.cpp b/src/engine/make1.cpp
index 1b1858af6..6a99f83b5 100644
--- a/src/engine/make1.cpp
+++ b/src/engine/make1.cpp
@@ -708,8 +708,8 @@ static void make1c( state const * const pState )
                      *    considered built before the additional MAKE1A state
                      *    processing even got a chance to start.
                      */
-                    make0( t->includes, t->parents->target, 0, 0, 0, t->includes
-                        );
+                    make0( t->includes, t->parents ? t->parents->target : 0, 0,
+                        0, 0, t->includes );
                     /* Link the old includes on to make sure that it gets
                      * cleaned up correctly.
                      */

With the patch applied the above for loop is absolutely stable. Even 10000 iterations do not trigger a single crash.

However I am not really sure if simply passing 0 to make0() if t->parents turns out to be 0 is the correct way to address the problem.