1 |
On 17-07-2009 09:01:14 +0200, Markus Duft wrote: |
2 |
> On Thu, 2009-07-16 at 17:43 +0200, Fabian Groffen wrote: |
3 |
> > Hi Markus, |
4 |
> > |
5 |
> > On 16-07-2009 17:02:03 +0200, Markus Duft wrote: |
6 |
> > > i just (seemingly) finished preserve-libs support for x86-interix. |
7 |
> > > attached is the svn diff against the modified tree, including the patch |
8 |
> > > and the epatch... |
9 |
> > > |
10 |
> > > ok to commit, or do you want to integrate this into the prefix portage |
11 |
> > > svn branch? |
12 |
> > |
13 |
> > It looks good to me, so I can commit it to the prefix branch. Just one |
14 |
> > question (two actually): |
15 |
> > - is it really a good idea to hardcode the path to objdump? (don't we |
16 |
> > have symlinks to that stuff in our Prefix?) |
17 |
> |
18 |
> hm. the problem is the following: prefix has it's own objdump (binutils |
19 |
> build fine...), but that one does not know about the "microsoft special |
20 |
> hacks" about shared libraries (RPATH, SONAME, NEEDED), so exactly what |
21 |
> i'm looking for. /opt/gcc.3.3/bin/objdump is present in all interix |
22 |
> installations which install the GNU SDK (which is all installations that |
23 |
> want to build software :)). |
24 |
> |
25 |
> Alternatively i could remove objdump from binutils, and link to |
26 |
> the /opt/gcc.3.3/bin one, if that feels better. (i can't just use native |
27 |
> binutils, because "ar" at least is somewhat unusable with c++). |
28 |
> |
29 |
> > - you mention PECOFF is actually more closer, but that winnt uses |
30 |
> > different stuff but the same file; how is that going to work? does |
31 |
> > Interix' objdump know something about .dll files or something? |
32 |
> |
33 |
> They share the basic file format ("container"), but have completely |
34 |
> different contents. i'll have to do the same again (so a separate |
35 |
> LinkageMapWinnt class), since objdump is not able to give usefull infos |
36 |
> on .dll's... :( i will, for that purpose, have to write a small utility |
37 |
> i can distribute with parity, which uses it's internal facilities to |
38 |
> provide the required informations - thats not too much work, and parity |
39 |
> has to be there anyway when building winnt binaries. |
40 |
|
41 |
If the windows stuff will use another format, then I don't understand |
42 |
the decision to call it ITX instead of PECOFF. How about writing a |
43 |
scan{elf,macho} brother that does the job? e.g. scanpecoff and/or |
44 |
scandll? |
45 |
That way we don't need objdump at all, but use our own tool, that we use |
46 |
furteron on the process as well, reducing your work to parse the |
47 |
output of objdump in the bash code. |
48 |
|
49 |
> Anything too hackish to make it into portage? ;) i guess it's the most |
50 |
> clean way i can think of ATM... |
51 |
|
52 |
|
53 |
-- |
54 |
Fabian Groffen |
55 |
Gentoo on a different level |