From: sfuld@alumni.cmu.edu.invalid   
      
   On 2/11/2026 2:12 PM, John Levine wrote:   
   > According to Terje Mathisen  :   
   >>> We knew it was possible to do runtime code generation, since it's an   
   >>> ancient technique widely used by sort programs to speed up the inner   
   >>> comparison loop, but we had enough trouble getting our programs to   
   >>> work the normal way.   
   >>   
   >> 40 years ago my own sort functions/programs would instead either extract   
   >> the compound keys into a form that could be compared directly, or   
   >> re-write (with a reversible transform) the data in place, so that it was   
   >> trivially comparable.   
   >>   
   >> The final option was to extract a short prefix (8-bytes) of the key data   
   >> and accept that the full sort would only deliver a partially sorted   
   >> results, so that I had to do a final sort pass over each chunk of equal   
   >> prefix.   
   >   
   > I'm pretty sure you'll find all of those in sort programs from the 1960s.   
   > Sorting used more computer time than anything else so they put a great   
   > deal of work into making it fast.   
   >   
   > See, for example:   
   >   
   > https://bitsavers.org/pdf/ibm/360/os/R01-08/C28-6543-3_Sort_Merge_Feb67.pdf   
      
   Yup.  And note that having a faster sort program than the standard IBM   
   one made the company Syncsort a lot of money.   
      
   https://en.wikipedia.org/wiki/Precisely_(company)   
      
   --   
     - Stephen Fuld   
   (e-mail address disguised to prevent spam)   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)