Re: removing the 'make install'-->'make all' dependency

[Alexander Malmberg]

Nicola Pero wrote:
> 
> Hi.
> 
> I'm considering removing the dependency of 'make install' from 'make all'.

I understand the reasons to remove this, but personally, I've always
thought that this was a Good Thing (and I've wondered why just about
every other program and library got it wrong :).

[snip]
>  - compiling as superuser usually creates object files which the normal
> user has not enough permissions to touch.  Typically, you type 'make
> install' as superuser, that compiles a .m file into an .o file and links.
> Next time you run 'make' as a normal user though, building will fail
> because the normal user can't delete the .o file owned by the superuser!
> Annoying, and might be madly confusing for newbies.

Unless I'm mistaken, the user only needs write permission on the
directory the .o file is in to remove it.

[snip]
>  - there is a huge overhead in the development process; when developing,
> you usually do:
> 
>  <typing typing coding coding>
>  make
>  su -c 'make install'
>  <checking that it works>
> 
> Now the fact that 'make install' always runs 'make all' a second time
> causes the process to take up something like 50% more time; without the
> dependency, a typical build/install session could be much faster.

But you still have all the overhead of having to run su and make twice.
In my setup, I have a group of "trusted" users who can install stuff
(except where it'd affect root), so I only have to run "make install" as
myself to compile and install everything. Keeping root out of the build
process seems like a Good Thing, anyway (especially when hacking
makefiles; safer to catch "mishaps" as a normal user :).

> Of course we loose the fact that if you've not compiled something, 'make
> install' will detect it and complete the compilation.  But in my
> experience, because 'make install' is run as root and not as the standard
> user, when this 'automatical update of object files' happens is usually
> more confusing/annoying than helpful ... it seems to me that the average
> developer is actually paying (in terms of additional build time and
> additional complications) a lot for this feature,

I think it's very useful, and my setup works great for working on
GNUstep, but I wouldn't call it "average", so I can see why you'd want
to change this.

- Alexander Malmberg