lua-users home
lua-l archive

[Date Prev][Date Next][Thread Prev][Thread Next] [Date Index] [Thread Index]

On Wed, 23 Jul 2014 20:21:42 +0800
David Crayford <> wrote:

> I've recently been investigating Lua structure packing libraries. 
> vstruct is a thing of beauty as it allows me to layout a format
> string that looks like a structure,
> and it supports names, tables etc. The downside is that compared to 
> struct it's very slow. struct is orders of magnitude faster but it's 
> interface is not so nice. It brings
> me to tears to choose a library for speed over beauty but performance
> is important to my project.
> How easy would it be to rewrite certain parts of vstruct in C or
> maybe even front end struct with the same semantics as vstruct? The
> format string below is what
> I've knocked up for vstruct.
> local fmtstr =
> [[
>    request:  s4
>    key:{
>      department: s1
>      employee_no: s5
>    }
>    name: s20
>    addr1: s20
>    addr2: s20
>    addr3: s20
>    phone_no: s8
>    timestamp: s8
>    salary: s8
>    start_date: s8
>    remarks: s32
>    msg: s60
>    upd_ind: {
>      name: s1
>      addr1: s1
>      addr2: s1
>      addr3: s1
>      phone_no: s1
>      salary: s1
>      start_date: s1
>      remarks: s1
>    }
>    browseeq: s8
>    update_ind: s1
>    delete_ind: s1
>    add_ind: s1
>    error_ind: s1
>    browse_ind: s1
>    dept_ind: s1
>    count: s2
>    total: s2
>    item_no: s2
>    sbtot: s2
>    sbnum: s2
>    lineout: { 10 * {
>      filler1: s6
>      employee_no: s5
>      filler2: s3
>      employee_name: s20
>      filler3: s3
>      employee_phone: s8
>      filler4: s34
>    }}
> ]]

It's a process. You need to find some way of profiling vstruct to see
where it's burning up all that time,  and fix that one part. If it's
several order of magnitude slower, I'm betting that you'll find one
little tiny part that's written stupidly (hey, I've written some pretty
stupid code myself), and you can easily fix that one little spot to
perform well, and because it's the bottleneck, the throughput of the
whole will rise proportionally to the bottleneck you cleared.

Rewriting the whole thing in C won't help that much if you translate
the bad algorithm. I'd first find out where all that time is getting


Steve Litt                *
Troubleshooting Training  *  Human Performance