| 1 |
On 02/14/2013 07:53 AM, Fabian Groffen wrote: |
| 2 |
> On 14-02-2013 00:15:44 +0100, Ruud Koolen wrote: |
| 3 |
>> I have been working on exactly that; I have a proof of concept implementation |
| 4 |
>> [1] of a version of prefix that uses its own libc like this, which has been |
| 5 |
>> tested more-or-less successfully by a handful of people. I'm planning on |
| 6 |
>> developing it into a real implementation that could potentially be merged as |
| 7 |
>> an option into prefix, but work on that has not yet started. |
| 8 |
> |
| 9 |
> Great work! |
| 10 |
> |
| 11 |
>> The reasons it would be a good thing include improved independence from the |
| 12 |
>> host system, leading to reduced platform-specific breakage and |
| 13 |
>> platform-specific hacks. However, it cannot completely replace the current |
| 14 |
>> structure, as it relies on having a libc implementation available for every |
| 15 |
>> platform supported by prefix, which is not the case. Hence, if merged, it |
| 16 |
>> will always be as an alternative to the native libc (and of course, having |
| 17 |
>> parallel infrastructure for both models has its own downsides). |
| 18 |
> |
| 19 |
> Given that the libc is basically only a problem on Linux (due to the |
| 20 |
> many vendors around, and all doing it slightly different, as opposed to |
| 21 |
> UNIX systems or derivatives that have a very stable env for each |
| 22 |
> "release") I would like to see a solution that we can "turn on" for |
| 23 |
> Linux only, to get rid of all those glibc-is-too-old bugs and |
| 24 |
> workarounds. |
| 25 |
|
| 26 |
Another reason to avoid glibc on proprietary Unix systems is supportability: |
| 27 |
Unix vendors unlikely would support such setups, so even when there's a bug |
| 28 |
in their kernel, we'd be on our own. Same basically stands for binutils. |
| 29 |
|
| 30 |
Example: Currently I'm working with AIX kernel developers to nail down why a |
| 31 |
(quite large) C++ executable built with (my old stable snapshot of) Prefix' |
| 32 |
gcc-4.2.4 does not start on aix7.1 - it does not even enter main(), while it |
| 33 |
works on aix5.3 and aix6.1. They tried to close the report when they heard of |
| 34 |
gcc, but continued when they were told that gcc just generates assembler code |
| 35 |
subsequently processed by their native binutils. |
| 36 |
|
| 37 |
/haubi/ |
Archives