On Sunday, 26 May 2013 at 12:59:27 UTC, Walter Bright wrote:
On 5/26/2013 4:03 AM, Paulo Pinto wrote:
Am 26.05.2013 09:27, schrieb Walter Bright:
That would imply that statically removing null pointers would
halve
total debugging time. Is there any evidence of that?
Well at least for Android, yes there is a proof.
http://blog.bugsense.com/post/49924755479/bigdata-in-motion-building-a-real-time-android
You're assuming that "all program bugs exhibit themselves as a
crash". This is quite false.
True, clearly the claim of "halve" is false, but I think we can
all agree from this data that null pointers are a frequent cause
of crashes, for some value of "frequent".
Given that your primary area of work is in compilers, you will
have a skewed perspective of what types of bugs are common.
Compilers typically allocate a lot, to build up the AST, perform
semantic analysis, emit code, then shutdown. There isn't much
opportunity for null pointer dereferences in this kind of program
because pointers are rarely set to null after they have been
initialised.
In contrast, interactive applications such as games and GUI apps
have many opportunities for null pointers. For example, GUI's
typically have a pointer/reference to the widget that currently
has focus, and may be null if no widget has focus. Everything
that uses that reference has to check for null, and it's easy to
forget. It's also difficult to unit test GUIs. A language that
statically enforces the programmer to check for null would help
here.
