Conference smurf::buildhelp

740.0. "Re: w/the qar, this time...spec. of include files... where are they, etc." by SMURF::FILTER (Automatic Posting Software - mail to flume::puck) Thu May 26 1994 18:53

Date Of Receipt: 	17-MAY-1994 12:18:13.71
From: 	FLUME::jmcg "Jim McGinness"
To: 	flume::dutile
CC: 	flume::buildhelp
Subj: 	Re:  w/the qar, this time...spec. of include files... where are they, etc.

What ends up in the export area and what ends up in the output area
were never explicitly expected to be the same, as far as I know, but
it would have made sense.  The export area is used in leiu of the
host's /usr/include files and libraries during an ODE build.

Since this hasn't been under any sort of regular control, there's not
much resemblance today between the structure of the output area and
the structure of the export area.  As best I can infer what the rule
should be, no code we write should use the construction "sys/machine/"
or "sys/alpha/" in the name of a header file.  The /usr/include/sys
directory in the export tree does not contain any subdirs.  This means
the fix for Al Carito's problem is to find and change the bad header
file reference, not to add links to the export tree.  ("Just because
something is accessible by a path doesn't mean that path is its name.")

I'm copying "buildhelp" on this to see if there are contrary views.

	-- jmcg

===========================================================================

From dutile Tue May 17 10:58:39 1994
Cc: dutile
Subject: w/the qar, this time...spec. of include files... where are they, etc.  
Date: Tue, 17 May 94 10:58:32 -0400
From: "Don Dutile, UNIX Systems Hardware Support" <dutile>


Jim,
	I have the following/attached qar.  On someone's
recommendation, I'm contacting you for advice as to what
is the proper fix/resolution.
	basically, should the export area (in a sandbox/
backing-tree) be a direct copy/reflection of the output area
on a system?  a secondary/curiosity question: is there 
a standard and/or documentation stating where the various 
include file paths are (on a system/output-area)?  
	finally, if the correct thing to do is to add
the necessary (directory) link to export/usr/include/sys/machine,
could you help (verify) the (Makefile) fix?  I think I figured
it out, but I want to double-check with someone who is
proficient in this space.  I ran this by Josh (ah, he's the
one who recommended I contact you...) and he recommended I
run it by you.
	advance thanks for the review.  btw, i dropped by to
discuss this one-on-one, but you weren't in, so i took the 
advice on your white-board -- send email.

/Don

==========================================================


Date: Mon, 9 May 1994 16:45:07 -0400
From: [email protected] (09-May-1994 1643)
To: [email protected]
Subject: High priority QAR 22945 from OSF_QAR database

 
QAR 22945 from OSF_QAR has been assigned to DUTILE
 
This QAR has been marked PUBLISH, so no immediate action is necessary.
Customers will be able to read this QAR even before it is answered.
 
QAR 22945  Version SID10275   9-MAY-1994 09:43:38.3

-- Submitted by --                      -- DIGITAL contact --
ALLEN CARITO                            FLUME::CARITO                           
3UC                                     DECWINDOWS, EMACS                       
FLUME,ABYSS                             HSC (2), VAXBI(3), RA (16)              
ZKOO3                                   
381-1527, ZKO3-3/X18                    

Attachments: NONE                      		Customer severity: H
Reproducible at will: Y

CPU        Memory     System device
VAX8800    60         RA        

Missing ./usr/include/sys/machine/hal_sysinfo.h

When building my code, I have not been successful in using finding 
./usr/include/sys/machine/hal_sysinfo.h in the export directory.
This has caused my builds to break.  The file does exist on a normally
installed machine, it does not exist in the export directory.  At
illustrated below:


Against goldos.nightly:

% ls -la 
/usr/sde/osf1/build/goldos.nightly/export/alpha/usr/include/machine/hal_sysinfo.
h
-rw-r--r--   1 8887     staff       6427 May  7 20:27 
/usr/sde/osf1/build/goldos.nightly/export/alpha/usr/include/machine/hal_sysinfo.
h

% ls -la 
/usr/sde/osf1/build/goldos.nightly/export/alpha/usr/include/sys/machine/hal_sysi
nfo.h
/usr/sde/osf1/build/goldos.nightly/export/alpha/usr/include/sys/machine/hal_sysi
nfo.h not found



Against a gold machine (X3.0-10  [Rev. 263.1]):

% ls -la /usr/include/sys/machine/hal_sysinfo.h
lrwxr-xr-x   1 root     bin           17 May  4 10:23 
/usr/include/sys/machine/hal_sysinfo.h -> hal/hal_sysinfo.h

% ls -la /usr/include/machine/hal_sysinfo.h
lrwxr-xr-x   1 root     bin           17 May  4 10:23 
/usr/include/machine/hal_sysinfo.h -> hal/hal_sysinfo.h

Date Of Receipt: 	17-MAY-1994 14:28:02.14
From: 	FLUME::jmcg "Jim McGinness"
To: 	flume::dutile
CC: 	flume::buildhelp, flume::gpt
Subj: 	Re: w/the qar, this time...spec. of include files... where are they,  etc.

You're right.  The link /usr/sys/include/sys/machine-> ../machine is
anomalous and should go away.  Michael Fairbrother added this back
in Feb 1992 and appears to have merged it in while distracted with
doing a large merge into Sterling.

This mistake was new to Sterling, so I think there's still time to
undo it.

I have a feeling that every single header file shipped in the
directory /usr/sys/include (as opposed to one of its subdirs)
is a mistake, but we've been making this error persistently.

	-- jmcg