I wanted to change this before rc2, but...

> But upon linking dynamically, gst becomes less stable, at
> least when working with FFI.

... can you expand on this?

With previous releases (e.g. 2.95c) I could build gst till the end, install it and launch it. With 2.95f I have ./cofigure && make failed:

./gst --no-user-files --kernel-dir="/home/ildar/tmp/RPM/BUILD/smalltalk-2.95f/kernel" -I "/home/ildar/tmp/RPM/BUILD/smalltalk-2.95f/gst.im" -iQ /dev/null
BlockClosure new "<0x40301398>" did not understand selector /var/tmp/RPM/BUILD/smalltalk-2.95f/kernel/ObjMemory.st:628: Aborted
(ip 8)BlockClosure>>#ifCurtailed:
(ip 10)Delay class>>#runDelayProcess
(ip 4)[] in Delay class>>#startDelayLoop
(ip 28)[] in Process>>#onBlock:at:suspend:
(ip 10) BlockClosure>>#on:do:
(ip 14)[] in Process>>#onBlock:at:suspend:
(ip 2) BlockClosure>>#ensure:
(ip 10)[] in Process>>#onBlock:at:suspend:
(ip 14)BlockContext class>>#fromClosure:parent:
make[2]: *** [gst.im] Aborted

The only change I made is in the _simple_ patch.

I can reproduce it.

Reduced test case (from arrays.st)

SortedCollection extend [
    SortedCollection class >> bug2 [
        ^##(#(8 6 9 10) asSortedCollection)
    ]

    bug: a [
        self sortFrom: lastOrdered + 1 to: lastIndex.
        ^(lastOrdered + 1 to: lastIndex) collect: [:each | a].
    ]
]

^SortedCollection bug2 bug: 5

Fixed with this simple patch:

--- orig/libgst/md-config.h
+++ mod/libgst/md-config.h
@@ -87,13 +87,7 @@
 # define REG_AVAILABILITY 0
 # define __DECL_REG1 __asm("%esi")
 # define __DECL_REG2 __asm("%edi")
-# if defined __APPLE__
-#  define __DECL_REG3 /* bug on Apple GCC? */
-# elif defined __PIC__ || defined __pic__
-#  define __DECL_REG3 __asm("%edx")   /* Don't conflict with GOT pointer... */
-# else
-#  define __DECL_REG3 __asm("%ebx")   /* ...but prefer a callee-save reg.  */
-# endif
+# define __DECL_REG3 /* no more caller-save regs if PIC is in use!  */
 #endif
 
 #if !defined(__DECL_REG1) && defined(PPC) || defined(_POWER) || defined(_IBMR2)

I have not yet removed the flag though.

I see what you mean with "FFI is less stable". GDBM tests fails, and on x86-64 actually I cannot even build gst. I think for 3.0 I'll go with static linking.

Doh, I know what's happening. It's enough to define NO_MALLOC_OVERRIDE to fix it... and a lot of code just becomes dead...

fixed in 2.95g