Continues r13011. The .warn files that 4glpc writes - which never reach the
build log - had the same class of problem in the two directories corecompile
reaches after tools/asql. Both are now free of diagnostics that GCC 14+ treats
as errors; a full clean build stays at exit 0.
tools/adbload2 (22 down to 0):
- process.ec and jtypes.c call runtime entry points with no prototype in scope.
Declared in a4gl_dbload_int.h rather than by including a4gl_libaubit4gl.h,
because that header defines A4GL_assertion as a two argument macro and
jtypes.c calls it with parenthesised arguments the preprocessor miscounts.
The block is guarded on _AUBIT_LIB_INCL_EXT_ so the translation units that do
include the header keep using its versions and do not get redundant
declarations.
- Forward declarations for the functions defined further down process.ec, and
for those in jtypes.c, which is #included near the bottom of it - so a
declaration in jtypes.c itself comes too late for callers above.
- Declaring A4GL_assertion turned up three calls that never matched it: one
passing no message at all, and two passing a printf-style format plus an
argument to a function that takes (int, char *) and does no formatting. They
now build the message with snprintf first.
- Three A4GL_assertion calls passed dbload->f->delim, a char *, as the int
condition; they mean "is it set", so they now say != NULL.
tools/asql/runforms (3 down to 0):
- A4GL_isnull, A4GL_read_columns and aclfgl_a4gl_get_info were declared only in
a4gl_libaubit4gl.h, which generated 4gl code does not include - it takes the
a4gl_incl_4glhdr.h path. Moved to a4gl_incl_4gldef.h, which is where CLAUDE.md
says shared declarations belong. Signatures checked against stack.c,
sql_common.c and helper.c rather than copied from the header.
- dbi.4gl passes &lv_dtype and &lv_size to A4GL_read_columns, which writes ints,
but 4gl integers are longs: the top four bytes were left uninitialised on
LP64. Same bug and same fix as the sqgetdbs calls in r13011.
Also in lib/libaubit4gl/ui.c: A4GL_open_gui_form declared its last parameter as
"void (*handler_c (int a, int b))", ie a function returning void *, while every
caller passes int (*)(void *, void *) and the function it forwards to,
A4GL_open_gui_form_internal, takes a plain void * - as does handler_e beside it.
Made it void *. That declaration moved to a4gl_incl_4gldef.h too, since the
callers are generated 4gl code.
Not done: glade/ide and glade/ide/sqleditor still have 32, mostly generated 4gl
code calling aclfgl_* functions from sibling modules without prototypes. glade
is a separate target from corecompile, built only when enable_minimal is not
set, so it does not block the core build.
One thing tried and reverted: declaring compat.c's bare trim() in
a4gl_libaubit4gl.h. The name is too common - lib/liblogical/txt/process has its
own static trim() - and it broke the build. Callers that want the library one
declare it themselves; there is now a comment in the header saying so.
