[Firebird-test] What to do with TCS

[John B. <bel...@cs...>]

Hi,
	I've been toying with the problem of replacing TCS while still 
being able to use all the existing tests.  Dejagnu was proposed as a 
possible new testing system, but that doesn't address TCS 
compatibility.  So I spent some time re-implementing TCS in Tcl (the 
language dejagnu uses).  My current implementation runs off the flat 
files generated from Frank S-G's TCS dump scripts, all flat files.

TCS/Tcl appears to function correctly with the C_SQL_JOIN, C_SQL_PRED, 
CF_ISQL, and QA_PROCS_ISQL series.  There are still a number of issues 
that need to be ironed out, but I figured I tell everyone what I've been 
up to.  This brings me to the issue at hand, what system do we want to 
use for our regression testing?

I've seen two ideas:

1.  A modified TCS/c that uses flat files.

2.  Dejagnu (Tcl/expect based).

The TCS/Tcl work I've done shows it is feasible to use (2), and that is 
what I'm advocating.  Tcl as a testing language gives us precise control 
over the interactions between the programs being tested, something that 
is very difficult in c.  For example Mark wrote a small script to test 
limit.  The script is able to count the number of rows returned from a 
query.  Using these sort of techniques is much easier than trying to 
implement them in c or manually inspecting the result of the queries to 
ensure the diff file is correct.

There are a number of other advantages, including the use of flat files 
and the fact dejagnu is a fairly standard testing tool.  Mark was saying 
the Postgres and mySQL test suites are written in dejagnu.  It may be 
possible to use some of them with FB.

What do other people think?  It would be nice to decide on a direction 
so we can make more focused progress.

Thanks.

-John