Question re "--prefix". DESTDIR, and actual installation dir

Jonathan Wakely
Sun Feb 4 23:57:00 GMT 2018

On 3 February 2018 at 12:42, Gordon McConnell
<> wrote:
> Hi all,
> I may be looking in all the wrong places, but so far I have not seen any information clarifying the relationships amongst:
>   - the path specified via “--prefix” when I run configure
>   - the path that I may specify via “DESTDIR=”
>   - the actual path where the compiler (or binutils etc. which use the same mechanisms) may end up
> and what (if any) problems may exist if the final installation path is not the same as the prefix.

No problems will occur.

> Consider the scenario:
>   1. I build the compiler, configured with a prefix of /tools/GCC/<version>
>   2. For the install step, I run that with a DESTDIR setting of $HOME/temp_GCC_installations/<version>  and use that intermediate location to do some testing prior to the intended final installation under /tools/GCC
>   3. When it comes to do the final installation, we get it in the expected /tools/GCC location at SOME of our sites, but one site reports that /tools is full, so they have installed the compiler under /tools2/GCC instead
> Does the /tools2 location pose any potential problems due to it not matching the prefix that the compilers were built with ?


> I have so far not seen any problems with the compiler being installed in a location that differs from the prefix – and the fact that I can run a battery of test from the intermediate DESTDIR location before the final installation has happened, suggests that the compiler seems perfectly fine being somewhere other than the prefix location.  However, if the location where they will be installed doesn’t actually matter, then why do gcc / g++ etc. carry with them the information on where they were intended to go into /tools/GCC (gcc -v tells you what prefix it was configured with).

They record the entire configure command. If that includes a --prefix
then that gets recorded. It doesn't mean the installed files can't
live elsewhere.

> Thanks in advance for any enlightenment that you can send my way.
>    Gordon

More information about the Gcc-help mailing list