| Directory | Changes | Lines of Code | Lines per Change |
|---|---|---|---|
| Totals | 40 (100.0%) | 695 (100.0%) | 17.3 |
| proxy/http2/ | 10 (25.0%) | 287 (41.3%) | 28.7 |
| / | 7 (17.5%) | 197 (28.3%) | 28.1 |
| proxy/ | 10 (25.0%) | 54 (7.8%) | 5.4 |
| libinktomi++/ | 5 (12.5%) | 52 (7.5%) | 10.4 |
| proxy/api/ts/ | 1 (2.5%) | 37 (5.3%) | 37.0 |
| librecords/ | 3 (7.5%) | 36 (5.2%) | 12.0 |
| iocore/cache/ | 2 (5.0%) | 30 (4.3%) | 15.0 |
| iocore/dns/ | 1 (2.5%) | 1 (0.1%) | 1.0 |
| build/ | 1 (2.5%) | 1 (0.1%) | 1.0 |
fixes TS-635
2 lines of code changed in 1 file:
I need to learn to copy and paste better. invalid cast
1 lines of code changed in 1 file:
these don't work under Sun Studio on the buildbot
3 lines of code changed in 2 files:
this addresses TS-578
4 lines of code changed in 1 file:
TS-533 stack tracks on ARM, this needs review
8 lines of code changed in 1 file:
TS-533 and TS-135
This is a sad day of defeat. Not my defeat, but more a collective human defeat.
Question: "Chips fabricated today don't have 64bit atomic primitives?"
Answer: "Be sad."
The ARM box we're working on (armv5tejl) doesn't support any 64bit primitives.
This means we need a method of using a global (yes, giant lock of death) to
protect modifications of arbitrary 64bit integers in process space. We could
make this less contentious by allocating pagesize/8 mutexs and then protecting
an int64 based on its page offset. Instead, I think we should mobilize to
burn these architectures to the ground and use public embarrassment to fix
future instruction sets.
If another platform has this issue, we'll want to change the define to:
TS_ARCHITECTURE_LACKS_64BIT_INSTRUCTIONS and turn on the global death lock
based on that.
This does not change performance on any other platform -- it's compile time
capital punishment.
41 lines of code changed in 2 files:
TS-533 ARM needs to sue the i386 version of the ink_queue stuff
3 lines of code changed in 2 files:
TS-533
No idea why this complains about this.
gcc-4.3.3 results in "error: NULL used in arithmetic" on ARM.
for each of these. This too should simply a redundant cast
on platforms that don't present with this illness.
3 lines of code changed in 1 file:
I get a warning (-Werror error) for comparison of integers with different signedness.
This should be safe everywhere.
4 lines of code changed in 3 files:
TS-525 make this work with bourne shell -- it was bash only as it was
3 lines of code changed in 1 file:
make the tool a bit more friendly (output)
28 lines of code changed in 1 file:
we can't status_get if the source is SOURCE_NONE.. so don't
2 lines of code changed in 1 file:
add a license statement TS-512
16 lines of code changed in 1 file:
TS-512 it's neat that his is a power of two... tsxs to help compile modules
148 lines of code changed in 3 files:
TS-509 simple, brute force (and efficient) status code stats counters
274 lines of code changed in 3 files:
TS-508
Generalize RecDumpRecordsHt to use RecDumpRecords which is a callback/map pattern.
Map this into InkAPI with the new
INKRecordDump(INKRecordType rec_type, INKRecordDumpCb callback, void *edata)
function. This doesn't change any APIs, introduces one and should cause no
functional changes in the product.
Documentation of the INKRecordDump function is still required in ts.h
79 lines of code changed in 5 files:
This should be defined for everyone. fixed build failure on linux
2 lines of code changed in 1 file:
we actually need the 'compressBound' feature in zlib, so check for that. If no zlib, we have no uLongf... we expect a uint32 anyway.
4 lines of code changed in 2 files:
Two prototypes (different even) is a bit sloppy. Pull it to a header and
be consistent. Also, some pedantic compilers treat signatures funny on
char buffer[32] arguments (different from caller and callee) so lets go more
traditional C and do (int length, char *buffer). This should be clean and
reusable.
41 lines of code changed in 4 files:
This is used in an array size specification and this should be const
1 lines of code changed in 1 file:
(3 more)