On Apr 20, 2008, at 7:09 PM, Bryan Blackburn wrote:
[about the Leopard problem where some ports fail to install properly
with a seemingly endless variety of bizarre error messages if
installing them causes a dependency to be installed first, but the
port succeeds if you try the install a second time (because now it
has no dependencies to install)]
Actually, it has something to do with the way the environment is
setup and cleaned between stages within MacPorts. I've been trying
to figure out whether this is a tcl bug, or something specific to
the Mac OS X install of 8.4.7 (as of 10.5.2), since I can't make it
happen with 8.5.2.
The basic issue is best demonstrated with a bit of code (first with
the Mac-default 8.4.7):
% puts $env(CC)
can't read "env(CC)": no such variable
% set env(CC) "gcc"
gcc
% puts $env(CC)
gcc
% array unset env *
% puts $env(CC)
% puts [array get env]
... CC {} ...
Note how CC in the environment is now blank, but still exists,
after the 'array unset env *' (and if you run this yourself, you'll
see all the environment variables in that last command are {}).
Confirmed with Apple-provided tcl 8.4.7 on 10.5.2.
What it should do, and did (I believe) on 10.4 and does with 8.5.2:
% puts $env(CC)
can't read "env(CC)": no such variable
% set env(CC) "gcc"
gcc
% puts $env(CC)
gcc
% array unset env *
% puts $env(CC)
can't read "env(CC)": no such variable
% puts [array get env]
Here, after 'array unset env *' CC is no longer set in the
environment. Which is how it should work. Note also, the last
command doesn't show all env vars as empty but nothing there.
Confirmed with Apple-provided tcl 8.4.7 on 10.4.11.
Confirmed with MacPorts tcl 8.5.2 on 10.4.11.
Confirmed with MacPorts tcl 8.5.2 on 10.5.2.
We can't use 'array unset env' as that is documented to disconnect
Tcl from the environment, and some messing around hasn't shown a
clean way to work around this. Maybe a Tcl expert is needed, and
determining if this is Mac-specific or 8.4.7; of course, either
way, we have to deal with it.
FYI, this is done in portutil.tcl (base/src/port1.0/) in proc
command_exec.
Thanks for the analysis! I added a link to your explanation to http://
trac.macports.org/wiki/LeopardProblems
Do we really not have a macports base ticket for this issue yet? I
couldn't find one but maybe I'm searching wrong. We need one so that
we can mark all the other port bugs that are cropping up as
duplicates of it.
_______________________________________________
macports-users mailing list
[email protected]
http://lists.macosforge.org/mailman/listinfo/macports-users
