Skip to content

Five dynamically-loaded GTK/GDK functions are called through a wrong function-pointer type #3533

Description

@GreenFuze

Several GTK/GDK entry points are loaded at run time and called through a function-pointer type the stub declares itself. The cast tells the compiler what the signature is, so nothing is left to check against GTK's header — and five of them disagree with it.

Three read a return value from a function GTK declares void:

/* bundles/org.eclipse.swt/Eclipse SWT PI/gtk/library/gtk4.c:1127 — gtk_color_dialog_set_title */
rc = (jlong)((jlong (CALLING_CONVENTION*)(jlong, char *))fp)(arg0, (char *)lparg1);
/* gtk/gtkcolordialog.h */  void gtk_color_dialog_set_title (GtkColorDialog *self, const char *title);
/* gtk/gtkfontdialog.h   */ void gtk_font_dialog_set_modal  (GtkFontDialog  *self, gboolean modal);
/* gtk/gtkfontdialog.h   */ void gtk_font_dialog_set_title  (GtkFontDialog  *self, const char *title);

Same shape at gtk4.c:1887 (gtk_font_dialog_set_modal) and gtk4.c:1909 (gtk_font_dialog_set_title).

Two more disagree about the parameters:

/* bundles/org.eclipse.swt/Eclipse SWT PI/gtk/library/os.c:2494 */
rc = (jlong)((jlong (CALLING_CONVENTION*)(jint, jint))fp)(arg0, arg1);
/* gdk/gdktoplevellayout.h */ GdkToplevelLayout *gdk_toplevel_layout_new (void);

/* os.c:2514 */
rc = (jboolean)((jboolean (CALLING_CONVENTION*)(GdkToplevel *, jint, jint, GdkToplevelLayout *))fp)((GdkToplevel *)arg0, arg1, arg2, (GdkToplevelLayout *)arg3);
/* gdk/gdktoplevel.h */ void gdk_toplevel_present (GdkToplevel *toplevel, GdkToplevelLayout *layout);

What it can cause

ColorDialog.java:143 and FontDialog.java:171-172 call the first three, so that path runs. Java discards the value, so nothing goes wrong today — but reading the return of a void function is undefined behaviour, and it stops being harmless the moment anything uses rc.

For the other two I could find only the native declarations in GDK.java and no caller, so I believe they are unreachable — please treat that as my reading rather than a claim. If gdk_toplevel_present were called, GDK would take its second argument, a jint here, and use it as a GdkToplevelLayout *.

No observed failure; nothing was built or run.

Fix

gtk_color_dialog_set_modal, the fourth member of the same family, is already right:

// GTK4.java:460
public static final native void gtk_color_dialog_set_modal(long self, boolean modal);
/* gtk4.c:1106 */ ((void (CALLING_CONVENTION*)(jlong, gboolean))fp)(arg0, (gboolean)arg1);

So the other three want the same change: longvoid in GTK4.java at :426, :432 and :444, then regenerate. Both call sites already ignore the result.

The two GDK declarations need the arguments GDK actually takes:

// GDK.java:1139 and :1199, as they would read after the change
public static final native void gdk_toplevel_present(long toplevel, long layout);
public static final native long gdk_toplevel_layout_new();

How this was found

By a research tool that compares each Java native declaration against the C declaration of the function its generated stub actually calls — here the case where the stub declares that signature itself, so no compiler sees both ends. Every line above was re-read on master (3e3b24b4) and against the GTK 4.23.2 headers.


Edited 2026-08-21: shortened. The claim, the evidence and the line numbers are unchanged.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions