The function `SET_OF__encode_sorted` stored the working element buffer
in a static variable. If called recursively (that is, when performing
encoding of a SET containing another SET) the outermost element buffer
is therefore clobbered by the innermost call.

This causes a SEGV on continuation of the outermost call.

Bug found during testing of [https://github.com/surevine/spiffing Spiffing],
fix verified with the same test suite.

In ASN.1 of 38.331 rev 15.2.1, module NR-InterNodeDefinitions
imports GapConfig from NR-RRC-Definition module. It should treat

  SetupRelease { GapConfig }

in module NR-InterNodeDefinitions the same as the one in
module NR-RRC-Definition.

For the following ASN.1 excerpt, the value of these Types will
be checked against its valid range in their corresponding
constraint functions.

DRB-Identity ::= INTEGER(1..32)
RSRP-Range   ::= INTEGER(0..97)
RSRQ-Range   ::= INTEGER(0..34)

Sometime it is convenient for application being aware of these
min and max values.

This commit generate the following macro defintions in
asn_constant.h :

\#define min_val_DRB_Identity (1)
\#define max_val_DRB_Identity (32)
\#define min_val_RSRP_Range (0)
\#define max_val_RSRP_Range (97)
\#define min_val_RSRQ_Range (0)
\#define max_val_RSRQ_Range (34)

When ActualParameter to a parameterized type is a Type, commit
59b9e705 will produce C struct types using ActualParameter,
which are conflict with the C struct types for these Types themselves.

For example, the following typedef has duplicate definition of
UCI_OnPUSCH_t :

typedef struct UCI_OnPUSCH {
	UCI_OnPUSCH_PR present;
	union UCI_OnPUSCH_u {
		NULL_t	 release;
		UCI_OnPUSCH_t	 setup;
	} choice;

	/* Context for parsing across buffer boundaries */
	asn_struct_ctx_t _asn_ctx;
} UCI_OnPUSCH_t;

With this commit, if ActualParameter is a Type, then prefix it
with Parameterized Type name.

typedef struct SetupRelease_UCI_OnPUSCH {
	SetupRelease_UCI_OnPUSCH_PR present;
	union SetupRelease_UCI_OnPUSCH_u {
		NULL_t	 release;
		UCI_OnPUSCH_t	 setup;
	} choice;

	/* Context for parsing across buffer boundaries */
	asn_struct_ctx_t _asn_ctx;
} SetupRelease_UCI_OnPUSCH_t;

As for ActualParameter is ObjectSet, used in S1AP ... etc, it doesn't
prefix generated typedefs for keeping names shorter.

This commit can replace item 3 of

https://github.com/vlm/asn1c/issues/282#issuecomment-390838895

which solve part of #282.

1. Record the de-referenced expression to newly added field ref_expr
   of strcut asn1p_ref_t.
2. In function asn1c_make_identifier(), not only check whehter there
   is name clash occurred for this asn1p_expr_t but also the referenced
   expression to determine whether an additional module name to be added.
3. Change signature of some functions and variables to eliminate warning
   messages when const type pointer, i.e. const asn1p_ref_t *, variables
   versus, non-const * type pointer, i.e. asn1p_ref_t *, varibles are
   used. E.g.:
   warning: assignment discards ‘const’ qualifier from pointer target type

Mimic what OAI's asnfix.c of generate_asn1 script does.

asn1c_save.c:238:5: error: this ‘for’ clause does not guard... [-Werror=misleading-indentation]
     for(int i = 0; i < argc; i++)
     ^~~
asn1c_save.c:240:2: note: ...this statement, but the latter is misleadingly indented as if it were guarded by the ‘for’
  safe_fprintf(mkf, "\n\n");
  ^~~~~~~~~~~~

../skeletons/INTEGER.c:510:60: error: comparison between signed and unsigned integer expressions [-Werror=sign-compare]
             if(specs && specs->field_unsigned && dec_value <= ULONG_MAX) {

This fixes the following undefined behaviour

converter-example.c:721:13: runtime error: null pointer passed as argument 2, which is declared to never be null
/usr/include/string.h:44:28: note: nonnull attribute specified here

Previously some uper_* function were called

When there were no any padding bits the uper and aper decoders where
trying to read some which results in read past the buffer end.

See https://github.com/mouse07410/asn1c/issues/33#issuecomment-374625610

ERROR: AddressSanitizer: heap-buffer-overflow on address 0x602000000315 at pc 0x000102ce9c15 bp 0x7fff5d20a8d0 sp 0x7fff5d20a8c8
READ of size 1 at 0x602000000315 thread T0
    #0 0x102ce9c14 in asn_get_few_bits asn_bit_data.c:132
    #1 0x102d878fd in aper_open_type_get_simple per_opentype.c:455
    #2 0x102d85dcc in aper_open_type_get per_opentype.c:514

On i686 the long is 4 bytes and when converted to uint64_t some big
unsigned values get converted to very big negative values

See https://github.com/vlm/asn1c/issues/185#issuecomment-372546490

On i686 the long is 4 bytes and the right shift with more then 32 bits
is probably an undefined behaviour

See https://github.com/vlm/asn1c/issues/185#issuecomment-372546490

Instead in two bytes the ENB-UE-S1AP-ID was encoded in three
For value 2 the correct encoding is `02 00 02 00`
and not `03 00 02 00 00`

And the MME-UE-S1AP-ID were encoded in 5 instead of 4

Files are from #185 and were renamed to match the Makefile's glob
pattern

bin/1.bin -> nextepc/sample-S1AP-PDU-1.aper

Thanks to @acetcom

Previously when decoding fails the memory was reset with memset after
casting the specifics to asn_CHOICE_specifics_t in order to get the size

The problem is that the OPEN TYPE could contains a different type not
derived from asn_CHOICE that does not have a struct_size member with a
valid size.

This fixes test failures in examples/sample.source.S1AP

This makes debugging and testing a lot more easier as skeleton files are
not copied and you don't need to execute `make regen` after every change
in the skeleton files.

You need to manually remove all skeleton files and run
`make regen-makefile` the first time after checking out this commit

Rename the existing files to match the glob used in the Makefile

To execute the tests run `make check` or `make check-aper`

and regenerate the the S1AP Makefile

According to Rec. ITU-T X.691 (08/2015)

11.5.3 Let "range" be defined as the integer value ("ub" – "lb" + 1), and let the value to be encoded be "n".

See also https://github.com/vlm/asn1c/issues/185#issuecomment-368745830

When generating code for multiple ASN.1 syntaxes that have clashing
names, we need to add a prefix in order to prevent clashes in the global
C symbol namespace.  Using the ASN1C_PREFIX environment variable and
this patch serves as a work-around to that.  All non-basic type names
as well as references to that type and source code + header file names
will be pre-fixed accordingly.

This is a re-implementation of this feature due to osmocom's
implementation version can not work on newest asn1c repository.

1. Copying and modifying ANY_decode_uper() and
ANY_encode_uper() to ANY_decode_aper() and
ANY_encode_aper().

2. Adding ANY_fromType_aper()/ANY_new_fromType_aper()/
ANY_to_type_aper().

Based on Velichkov's suggestion on comment :

https://github.com/vlm/asn1c/issues/185#issuecomment-341944947

For RANAP specification :

  RANAP-ELEMENTARY-PROCEDURE ::= CLASS {
    &InitiatingMessage,
    &SuccessfulOutcome          OPTIONAL,
    &UnsuccessfulOutcome        OPTIONAL,
    &Outcome                    OPTIONAL,
    &procedureCode              ProcedureCode  UNIQUE,
    &criticality                Criticality    DEFAULT ignore
  } WITH SYNTAX {
    INITIATING MESSAGE          &InitiatingMessage
    [SUCCESSFUL OUTCOME         &SuccessfulOutcome]
    [UNSUCCESSFUL OUTCOME       &UnsuccessfulOutcome]
    [OUTCOME                    &Outcome]
    PROCEDURE CODE              &procedureCode
    [CRITICALITY                &criticality]
  }

Not all instances of elementary procedure have all their
optional members defined.

For examples :

  iu-Release RANAP-ELEMENTARY-PROCEDURE ::= {
    INITIATING MESSAGE  Iu-ReleaseCommand
    SUCCESSFUL OUTCOME  Iu-ReleaseComplete
    PROCEDURE CODE      id-Iu-Release
    CRITICALITY         reject
  }

  rAB-Assignment RANAP-ELEMENTARY-PROCEDURE ::= {
    INITIATING MESSAGE	RAB-AssignmentRequest
    OUTCOME             RAB-AssignmentResponse
    PROCEDURE CODE      id-RAB-Assignment
    CRITICALITY         reject
  }

So there are empty cells generated in Outcome.c ... etc.

  static const asn_ioc_cell_t asn_IOS_RANAP_ELEMENTARY_PROCEDURES_1_rows[] = {
    { "&InitiatingMessage", aioc__type, &asn_DEF_Iu_ReleaseCommand },
    { "&SuccessfulOutcome", aioc__type, &asn_DEF_Iu_ReleaseComplete },
    { "&UnsuccessfulOutcome",  },
    { "&Outcome",  },
    { "&procedureCode", aioc__value, &asn_DEF_ProcedureCode, &asn_VAL_1_id_Iu_Release },
    { "&criticality", aioc__value, &asn_DEF_Criticality, &asn_VAL_1_reject },

    { "&InitiatingMessage", aioc__type, &asn_DEF_RAB_AssignmentRequest },
    { "&SuccessfulOutcome",  },
    { "&UnsuccessfulOutcome",  },
    { "&Outcome", aioc__type, &asn_DEF_RAB_AssignmentResponse },
    { "&procedureCode", aioc__value, &asn_DEF_ProcedureCode, &asn_VAL_49_id_RAB_Assignment },
    { "&criticality", aioc__value, &asn_DEF_Criticality, &asn_VAL_49_reject }
  }

However, in type_selector() functions, these undefined cell are still
be counted into presence_index value. This results in wrong behavior for
decoder.

This commit is to fix this problem.