[Toybox] lspci text output

[Felix Janda]

Isaac wrote:
> On Fri, Aug 02, 2013 at 10:35:30PM +0200, Felix Janda wrote:
> > Isaac wrote:
> > > On Sat, Jul 27, 2013 at 09:56:53AM +0200, Felix Janda wrote:
> > [...]
> > > > Anyway Isaac is doing (or planning to do) a major rewrite enabling lspci
> > > > to read the pciid database for human readable names of vendors and devices.
> > > 
> > > I have a question or two in regards to that.
> > > 1-mmap or fgets?
> > > It's a smallish file as data goes (currently~600 kb), and can be searched 
> > > in very little time if mmap()ed. 
> > > The alternative is an fgets() based loop, which could involve 
> > > reading over 11,000 entries, and decent performance would require 
> > > using a linked list for temporary storage, extracting and sorting the entries
> > > from the linked list, looping over the sorted entries and storing the text,
> > > then printing from the linked list (though you can print from the sorted data
> > > if you don't care about order).
> > > I'm thinking mmap sounds more practical.
> > 
> > I don't understand your argument. What would be the algorithm using mmap()?
> 
> With mmap, the time is small enough to search the whole file every time.
> But now that I try fgets, I think I overestimated the time it takes:
> $ time ./lookup-fgets "C 11" "80  " /usr/share/misc/pci.ids    
> 0: Signal processing controller : gnal processing controller
>     0m0.01s real     0m0.01s user     0m0.00s system
> 
> (5400 RPM HDD, Atom N270 locked to 800 MHz)
> When "Vendor" is "C **", device is 2 chars long, and I'm not sure what's a good way to handle that.
> See attached files for mmap (thanks to "jamesbond" at the Puppy forums) 
> and fgets examples.
> Usage is 
> lookup-fgets <vendor> <device> /usr/share/misc/pci.ids
> lookup <vendor> <device>
> lookup-fgets can be used with USB devices as well.

Thanks for the test programs. After clearing the caches
(echo 3 > /proc/sys/vm/drop_caches) it takes

    0m0.05s real     0m0.01s user     0m0.00s system

for me and nothing time(1)-measurable the second time
(720 RPM HDD, K8 CPU scaled to 1 GHz)

What are these classes used for? (Do we really need to care about them?)

Under usual circumstances the repeated searching shouldn't be noticeable.
It should be ok to use the simplest approach.

So one could even stay with the general current design? (Without the need to
keep a list of devices.)

> > How about the following approach using fgets? Say in a simplified version
> > we start out with a list of
> > 
> > struct foo {
> >   struct foo *next;
> >   struct bar *strs;
> >   char vendor[4], device[4];
> > }
> > 
> > , where
> > 
> > struct bar {
> >   struct foo *ref;
> >   char *vendor, *device;
> > }
> > 
> > , filled out from the /sys/ files. The fields strs are initialized to point
> > to elements (one for each device) of an array of struct bar and the ref fields
> > of the elements of this array should point back to the corresponding list
> > element. Then, sort the array of struct bars according to ref->vendor and
> > ref->device. Next, update ref->strs to point to the right element in the
> > sorted array. Now we can start reading the (sorted) pci database line by
> > line and fill out the other elements of the struct bar array. For the
> > printing we iterate through the list of struct foo.
>  
> That seems to make more sense than what I was thinking of, 
> if it really is needed (see time).

Yeah, quite some book keeping would be needed for the approach.

Felix

 1375558242.0