Re: [9fans] Regression testing

[Riddler <riddler876@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 2790 bytes --]

Hello,

Yes, I am envisioning automated tests.
I've built a limited set of tests for personal projects and for a company I
worked in for a while, so I have a rough idea of where to go.
I'll certainly look into system calls, seems a sensible place to start. Any
advice or pointers you (or anyone else) has would be greatly appreciated.
Also, I would quite like to see those fp tests if you would be so kind.

On Wed, Mar 26, 2014 at 4:25 PM, erik quanstrom <quanstro@quanstro.net>wrote:

> On Wed Mar 26 07:11:38 EDT 2014, riddler876@gmail.com wrote:
>
> > Hello,
> >
> > I would also be quite interested in helping if people are embarking on
> such
> > a project.
> >
> > I'm a Computer Science student and I've been using a raspberry pi trying
> to
> > learn more about plan 9. Creating regression tests could be a good way
> for
> > me to poke around the system learning how it works while still
> contributing
> > something useful to the community.
>
> i think this could be quite valuable.  at a minimum, you will learn alot.
>
> i assume that you mean automated tests.  many things like usb behavior
> on unplugging is hard to automate without some special equipment.
>
> imo, the easiest thing to automate would be the system calls.  there
> are very few of them.  i would first start by checking that they do
> reasonable
> things with arguments in the text, heap, stack, and various forms of
> invalid addresses, especially just above or just below valid addresses.
>
> it is also pretty straightforward to test a the kernel devices to make sure
> they all handle invalid file names/directory names well, deal with bad
> input to control files well, etc.
>
> (an idea that occurs to me while typing this is to have a special
> syscallmal
> for allocating things with system call lifetime.  it allows one to add the
> rule
> that everything syscallmal'd must be free'd on kernel exit.)
>
> another area that might not be as easy (as you'll have to dive into some
> obtuse stuff) would be floating point.  but this would be very helpful,
> because
> it has been a long time since serious effort was put into this.
> floating point should well behaved with the usual operations,
> and with standard library functions (and ape, too), with respect to the
> usual problem
> areas such as overflow, underflow,  -0., ±∞, and denormals, with all
> fcr (see getfcr(2)) settings, and with all architectures.  or at least
> amd64,
> rpi and something with emulated fp such as the rb.
>
> run-on sentences 'ᴙ' us.
>
> i'd be willing to give you some pointers on how to continue.  i have some
> scripts that i used to iron some of the worst bugs out of fp so that python
> could pass some regression tests.
>
> - erik
>
>

[-- Attachment #2: Type: text/html, Size: 3449 bytes --]