| Date Of Receipt: 29-JUN-1993 07:26:44.12
From: MINSRV::"[email protected]" "Michael D. Fairbrother UEG"
To: [email protected] (Jim McGinness)
CC: [email protected], [email protected], [email protected],
[email protected]
Subj: Re: What's your opinion on export area
Jim, your correct that there is a sequence problem, in that the
nightly build does not use the same process as an engineer would.
In fact I pointed out to Pete that he had missed the rule/list to
get the file into the exports area, but he has fixed this.
I also at the time spoke to Grant asking about how the nightly build
would know of such changes.
I previously sent mail to Pete stating that the ${MAKETOP} line was
bogus. When lastin comes back up I will try and rember what I did
to fix these Makefiles...
Oh, the rule-of-thumb is clearly stated, the build takes place in
passes, and files to get exported need to happen in an early pass
say EXP*_* or something... I belive that the nightly build does
a complete build as an engineer would at the end to catch this
kind of problem... so the fact that the 2 commands fail in one
part would mean that they should of still been build in the
final pass...(yuck)!
mdf.
|
| Date Of Receipt: 29-JUN-1993 10:28:06.76
From: FLUME::"[email protected]" "Grant Van Dyck"
To: [email protected] (Jim McGinness)
CC: [email protected], [email protected], [email protected]
Subj: Re: What's your opinion on export area
This is the 1st time we've had an occasion where commands depended
on datafiles in etc. In this case etc gets built early in the cycle
and the file in question is in place in exports/etc long before the commands
need it:
-rw-r--r-- 1 devbld staff 14977 Jun 28 22:51 builtin_types.dat
Even if this were not true, as MDF points out, there is a cleanup run
which would have caught this. Just so that everyone is sure that the script
is not the problem, I will restructure it so that this is done before
any command runs start.
Now, as to the problem. Mary Walker described the process as something
like one Makefile setting an environment variable that gets passed to another
Makefile (Ugggh), and in fact that part may not have been working. Additionally,
there are executables which must be built and run as part of the process
(much like awk and others), BUT HOST_CC or COMP_HOST_ROOT was not being set
and the objects are building using the backing tree libs and compilers.
Given the stated incompatibilites between MINOR and previous releases, and
the fact that all the build hosts are running 1.2 SSB, this is clearly wrong.
The rule is - if you have to run it during the build cycle, built it locally
else all bets are off.
-Grant
| The Common Agent build ran into a problem caused by not being able to find
| etc/eca/builtin_types.dat in the export area. This file was eventually
| exported there, but I haven't been able to track down how this happened
| out of sequence.
|
| The usr/sbin/internet_mom Makefile uses the construct
|
| ${MAKETOP}../../export/${target_machine}/etc/eca/builtin_types.dat
|
| which strikes me as being a suspicious expression because of dot-dotting
| out of MAKETOP.
|
| Also, because of the symptom in this case, is there some rule-of-thumb
| that says the exports area is complete at some point early in the build
| so the other passes or parallel phases of the build can rely on it?
|
| -- jmcg
|
|