Do all remaining changes necessary for a working build.
Add Makefile.am files in include/... subdirs.
Remove noinst_HEADERS directives from src/*/Makefile.am, but keep the headers
list to feed to move-asn1-header-files.sh.
Adjust all #includes in src/*_common.h and elsewhere. In hnbap_common.h,
separate the ASN.1 "primitive" headers from the others, and include them
without a subdir path, as before.
Add move-asn1-header-files.sh to do header file moving and sed'ding the include
statements. The file moving part is disabled until a later commit, to make
reading the diffs easier.
Call shell script from src/{hnbap,ranap,rua}/Makefile.am regen targets.
Add convenience regen target to src/Makefile.am, calling regen in the three subdirs.
This change is split over several commits to ease diff reading. Subsequent
commits show, in steps:
- the "unmoved" effect of sed,
- header moves,
- adjust build system and include statements.
in order to work around a bug in asn1c. When we keep the original
TBCD-STRING, the APER-encoded PLMNidentity always has an extra leading
length byte that the decoder doesn't expect.
When IMSI is a TBCD-STRING type, and TBCD-STRING is defined as OCTET
STRING, we end up encoding the IMSI the wrong way. I don't knwo why
that is, but changing it fixed the problem, as described below:
before this commit:
00 17 PeranentNAS-UE-ID
40 criticality ignore
0a (length)
00 presence = IMSI
08 BUG: why the additional length field?
46 23 91 34 70 77 80 f3 IMSI (643219430777083)
after this commit:
00 17 PeranentNAS-UE-ID
40 criticality ignore
09 (length)
50 presence = IMSI
46 23 91 34 70 77 80 f3 IMSI (643219430777083)
The definition of the above data types as per 3GPP specs results in a
SEQUENCE_OF() an anonymous structure, which is slightly inconvenient to
use. So let's split the SEQUENCE OF part and the actual definition of
the item in separate types.
This is not development, it is random trial and error hacking. I really
hate the fact that we have no useful asn.1 code generator and need to
work with hacks like asn1tostruct.py and asn1c without information
object classes :/
This commit is a one-day-long iteration of trial+error, manually editing
and adding the .asn source of RANAP until we get something that in the
end at least compiles and links. Do I trust the resulting code? No.
But we have no alternative :(
As asn1c cannot understand information object classes, we cannot compile
RANAP-PDU-Contents.asn but instead need to manually add the respective
infrmation elements to RANAP-PDU.asn.
This might need a lot of cleanup for out-of-source-tree builds and the
like, but let's not spend time on this now. The old Makefile also
didn't support that. But loosing the ability to regenerate the C source
is not an option either.
It seems that individual IEs contain nested containers, and
asn1c is not generating code for that unless we help it by some
hand-crafted additional definitions. *sigh*
Neels Hofmeyr
Neels Hofmeyr
Harald Welte
Alexander Huemer
Daniel Willmann