Files
heimdal/lib/asn1/README.template
2021-01-25 16:28:44 -06:00

164 lines
4.8 KiB
Bash

#!/bin/sh
size .libs/libasn1.dylib
size .libs/libasn1base.a | awk '{sum += $1} END {print sum}' | sed 's/^/TEXT baselib: /'
size .libs/asn1_*.o | awk '{sum += $1} END {print sum}' | sed 's/^/generated code stubs: /'
size *_asn1-template.o | awk '{sum += $1} END {print sum}' | sed 's/^/TEXT stubs: /'
exit 0
Notes about the template parser:
- assumption: code is large, tables smaller
- smaller tables and code, smaller cache footprint, better performance
- how to generate template based stubs:
make check asn1_compile_FLAGS=--template > log
or
./configure --enable-asn1-templating
- pretty much the same as the generate code, except uses tables instead of code
- much easier to extend! adding new encoding rules is just adding a few
functions to template.c, one set of length/encode/decode functions per ER,
so we could add OER/PER/XDR/GSER/JER with very little work outside that one
file and gen_template.c (to generate stub functions) and gen.c (to generate
declarations of those stub functions).
TODO:
- Fuzzing tests
- Performance testing
- ASN1_MALLOC_ENCODE() as a function, replaces encode_ and length_
- Fix SIZE constraits
- Proper implementation of `SET { ... }`
- Compact types that only contain on entry to not having a header.
Or maybe not. We can afford larger template footprint if we can get
more bang for that. For example, if we add type and member names to
the templates, we could have template interpreters that implement JER
(JSON Encoding Rules) and/or GSER (Generic String Encoding Rules),
and it will cost only the size of the new interpreters plust the
extra metadata (type and member names).
Perhaps we could rely on `dladdr()` to find the names of template
data structures and work out the type and member names from that?
But then, making more of those static / stripping them would save us
some of the cost of adding type and member names to the templates.
Using `dladdr()` might be a problem because it's not really portable,
and our implementation in lib/roken for WIN32 does not report a
symbol name, but there is a way to get that in WIN32
(see https://github.com/dlfcn-win32/dlfcn-win32/).
E.g., if we know the name of a template table as
"asn1_TBSCertificate_tag_version_28" then we can work out that it's
for the member named `version` in the type named `TBSCertificate`.
SIZE - Futher down is later generations of the template parser
code:
==================
__TEXT __DATA __OBJC others dec hex
462848 12288 0 323584 798720 c3000 (O2)
trivial types:
==================
__TEXT __DATA __OBJC others dec hex
446464 12288 0 323584 782336 bf000 (O2)
OPTIONAL
==================
__TEXT __DATA __OBJC others dec hex
425984 16384 0 323584 765952 bb000 (O2)
SEQ OF
==================
__TEXT __DATA __OBJC others dec hex
368640 32768 0 327680 729088 b2000 (O2)
348160 32768 0 327680 708608 ad000 (Os)
BOOLEAN
==================
339968 32768 0 327680 700416 ab000 (Os)
TYPE_EXTERNAL:
==================
331776 32768 0 327680 692224 a9000 (Os)
SET OF
==================
327680 32768 0 327680 688128 a8000 (Os)
TYPE_EXTERNAL everywhere
==================
__TEXT __DATA __OBJC others dec hex
167936 69632 0 327680 565248 8a000 (Os)
TAG uses ->ptr (header and trailer)
==================
229376 102400 0 421888 753664 b8000 (O0)
TAG uses ->ptr (header only)
==================
221184 77824 0 421888 720896 b0000 (O0)
BER support for octet string (not working)
==================
180224 73728 0 417792 671744 a4000 (O2)
CHOICE and BIT STRING missign
==================
__TEXT __DATA __OBJC others dec hex
172032 73728 0 417792 663552 a2000 (Os)
No accessor functions to global variable
==================
__TEXT __DATA __OBJC others dec hex
159744 73728 0 393216 626688 99000 (Os)
All types tables (except choice) (id still objects)
==================
__TEXT __DATA __OBJC others dec hex
167936 77824 0 421888 667648 a3000
base lib: 22820
__TEXT __DATA __OBJC others dec hex
==================
167936 77824 0 421888 667648 a3000 (Os)
baselib: 22820
generated code stubs: 41472
TEXT stubs: 112560
All types, id still objects
==================
__TEXT __DATA __OBJC others dec hex
155648 81920 0 430080 667648 a3000 (Os)
TEXT baselib: 23166
generated code stubs: 20796
TEXT stubs: 119891
All types, id still objects, dup compression
==================
__TEXT __DATA __OBJC others dec hex
143360 65536 0 376832 585728 8f000 (Os)
TEXT baselib: 23166
generated code stubs: 20796
TEXT stubs: 107147
All types, dup compression, id vars
==================
__TEXT __DATA __OBJC others dec hex
131072 65536 0 352256 548864 86000
TEXT baselib: 23166
generated code stubs: 7536
TEXT stubs: 107147