At 2:36 PM +0100 on 3/6/00, M. Uli Kusterer wrote:
>>I think dumping random data to random blocks and reading it out would
>>give it a better testing. Basicly, perform random actions to the file.
>>Include some erroneous ones (tryign to read non-existant blocks, for
>>example).
>
>Anthony,
>
> well, I'll certainly do testing with random data, too, but it'll really be
>helpful if we can test some real-life uses, too, as usually you test for
>everything except for, say the resource manager's limit of 2727 resources
>(total) per file...
Welll, that's why you really have to stress it. But yes, nothing beats
real-life tests for things such as performance testing.
However, you can generate _far_ more data _far_ quicker using random
methods than using real data.
Also we need to test things like:
open
start streaming write
abort()
open
set self up for signal in a few �s.
write
* signal interrupts write *
abort()
Basicly, we need to crash in the middle of XBF calls, both inside of
XBF and while streaming.
Losing the data being written is acceptable. Losing the entire block is
bad, but can be lived with. Losing the whole file or non-active blocks
is not acceptable. FreeCard must be able to die with grace, because it
will be happening, often, for the next few years.
>
>>Either that, or run a news server using XBF as a file system <g>.
>
> ohmygod. Anyhow, somebody's going to try anyway, right?
If it's faster than ext2, everyone will <g>.
