 Georg Hopp
|
931e2afae0
|
some more testing on the user class
|
12 years ago |
|
Georg Hopp
|
e58739bd63
|
add first user class and persistence experiments. refs #36
|
12 years ago |
|
Georg Hopp
|
7a319d3f76
|
initialize ref_count correctly
|
12 years ago |
|
Georg Hopp
|
7fa810c47e
|
add a ref_count to the memory segments...this way I could increase references and only insert the memory segment in the tree again if the ref_count is one.
|
12 years ago |
|
Georg Hopp
|
252cdbb0dc
|
now it works mostly as before...the only difference is that you really have to be logged in to see the rand value
|
12 years ago |
|
Georg Hopp
|
e69a4962bb
|
most stuff works...the session timeout will not be reset correctly within some answers, anyway it mostly works. refs #24
|
12 years ago |
|
Georg Hopp
|
79b346559a
|
code with abstraced application compiles again, but does not work correctly, start debugging. refs #24
|
12 years ago |
|
Georg Hopp
|
36612df0da
|
implemented an application class as well as an http adapter for it and use it to start application logic by http requests as well as creating a fitting repsonse. Not perfect, but a start. This code is not finish and will not work...in fact it won't even compile i think. refs #24
|
12 years ago |
|
Georg Hopp
|
04250fb43a
|
only mmap a file if its size is larger than 0 bytes
|
12 years ago |
|
Georg Hopp
|
9b7df5312b
|
fix wrong constant used in access
|
12 years ago |
|
Georg Hopp
|
6003df221e
|
fix error handling on SSL handles
|
12 years ago |
|
Georg Hopp
|
1bf8ec7c9c
|
use any value less than 0 as event done for ssl accept.
|
12 years ago |
|
Georg Hopp
|
16f7a42968
|
checked SSL support, the only thing that was really broken was the error handling while reading data... well maybe writing has also bugs, but it works.
|
12 years ago |
|
Georg Hopp
|
e0abf3ac91
|
now every date header is formatted in GMT. Commment: We still have a very weak Etag implementation.
|
12 years ago |
|
Georg Hopp
|
f93d09b5ca
|
now Date header will be formatted in GMT as requiered by HTTP protocol
|
12 years ago |
|
Georg Hopp
|
d9193d2e7e
|
now don't add a NULL asset to the hash.
|
12 years ago |
|
Georg Hopp
|
577b8ffa13
|
fix deactivation of SSL
|
12 years ago |
|
Georg Hopp
|
7ebd5a6a00
|
add subdir-objects to each Makefile.am
|
12 years ago |
Georg Hopp
|
d3e7244592
|
disabled SSL support for now, cause it seems that it crashed the whole thing right now.
|
12 years ago |
|
Georg Hopp
|
1249d85779
|
a fix in size of asset names and change in logging
|
12 years ago |
|
Georg Hopp
|
26da90232f
|
fix post order traversal used for destruction of existing tree
|
12 years ago |
|
Georg Hopp
|
b506318d72
|
fix the fix...
|
12 years ago |
|
Georg Hopp
|
a723c92c74
|
add a small test program for the hash code to find where it leaks memory.
|
12 years ago |
|
Georg Hopp
|
487c0cfb0b
|
add one missing delete for a node, thus fixing a memory leak in there.
|
12 years ago |
|
Georg Hopp
|
107896eb9d
|
now do all things within rbtree2 that are also done in rbtree
|
12 years ago |
|
Georg Hopp
|
5c10e78974
|
now use at minimum page size for an allocated memory segment, and also store this size. This should use the memory more effective.
|
12 years ago |
|
Georg Hopp
|
c8dc541c48
|
use now my own tree implementation as base for my hashes....sadly this thing is leaking memory again...on the other hand I workarounded the problem of a sometimes occuring infinite loop within connect.
|
12 years ago |
|
Georg Hopp
|
417b9f6d8f
|
most likely found a bug in the glibc tdelete implementation...don't have the mood of building a simple testcase for reporting. Maybe I do when I have implemented my own trees.
|
12 years ago |
|
Georg Hopp
|
ac5b20e9ff
|
as I currently have no idea whats the problem I first fixed all warnings...included the ldap warnings that where caused by the use of deprecated ldap_simple_bind_s and ldap_unbind_s.
|
12 years ago |
|
Georg Hopp
|
87bf963839
|
added forgotten file
|
12 years ago |
|
Georg Hopp
|
85a33ab3fd
|
I think the problem with keep-alive and concurrency is fixed...the handling in serverRun was still not ok for the non blocking sockets.
|
12 years ago |
|
Georg Hopp
|
56fdd4bd00
|
changed socket handling according to my definition...and make sockets nonblocking as the answer of poll is just a guess...
|
12 years ago |
|
Georg Hopp
|
817758272a
|
remove occurences of write circular buffer
|
12 years ago |
|
Georg Hopp
|
f9ef11ad76
|
fix cleanup of NULL asset_pool
|
12 years ago |
|
Georg Hopp
|
0e1ec66463
|
this is a huge one...removed cbuf for writing again.... This is not needed any more because I use memory mapped io for these. Additionally sanitised write error handling somewhat...anyway, under huge load I still observer unexpected connection closes. But maybe this is related to interrupt while reading...(reading is not handled very well right now)
|
12 years ago |
|
Georg Hopp
|
3289d758c9
|
now asset is by itself hashable...and the leaking issue seems to be fixed.
|
12 years ago |
|
Georg Hopp
|
d235aaef32
|
try to use memory mapped io for asset access, but this one leaks like a rotten barge
|
12 years ago |
|
Georg Hopp
|
acf4516b12
|
remove message type (PIPED|BUFFERD) because when I use mmap for file io I will always have a memory address, so its kind of buffered.
|
12 years ago |
|
Georg Hopp
|
d6cd2bbae9
|
added asset class and move mimetype handling in separate helper. Tested this with mmapfiletest2.c
|
12 years ago |
|
Georg Hopp
|
157b108220
|
added thought about memory mapped asset handling, as well as a small mmap file test program.
|
12 years ago |
|
Georg Hopp
|
5d2b27de1c
|
remove specialized asset loadings and use generic one.
|
12 years ago |
|
Georg Hopp
|
f8fd49938d
|
now allocate only a multiple of pagesize
|
12 years ago |
|
Georg Hopp
|
21abd36396
|
now sending non html data from one of the other folders also works
|
12 years ago |
|
Georg Hopp
|
623a062df7
|
add suppport for file extension based mime type detection.
|
12 years ago |
|
Georg Hopp
|
5c32329105
|
now every file accessible under assets/html will be delivered.
|
12 years ago |
|
Georg Hopp
|
157f48031c
|
now valgrind shows no more violations...but cleanup does not work correcly. The memory usage depends now on the amount of paralell connections...ab show approximately 10% to 20% performance improvement / and the code is far from optimal.
|
12 years ago |
|
Georg Hopp
|
55675eb50a
|
tree based memory management does not segfault anymore, but reusage does not seem to work, as well as free
|
12 years ago |
|
Georg Hopp
|
06266ef541
|
update todo and make my rbtree implementation use externally allocated elements.
|
12 years ago |
|
Georg Hopp
|
ef6ccfbcbe
|
rbtree insert and delete now as one function and working as expected...still has to be checked on sideeffects.
|
12 years ago |
|
Georg Hopp
|
52755ca9e9
|
realize that delete does not work correct.
|
12 years ago |
