Filters stream rebound #948
Conversation
|
Rebound of #946 |
|
Hmmm.... The |
|
@Therzok Tests are failing on AppVeyor. Looks like a memory issue. They pass when ran from Visual Studio, but launching |
|
w00t! The test runner is safe now (courtesy of @ammeep the CrashTamer™). |
|
One thing I'm going to add to this is some kind of smoother buffer growing algorithm (eg. if we're at 90% of asize, grow the buffer by 50%). Should have something working tomorrow |
| [Fact] | ||
| public void FilterStreamsAreCoherent() | ||
| { | ||
|
|
|
These changes look great. I'll pull this down and give it a test with a custom filter implementation of my own. |
|
/cc @ethomson |
|
I've taken this for a spin locally and it works just perfectly ✨ |
|
|
||
| public override void Write(byte[] buffer, int offset, int count) | ||
| { | ||
| AutoGrowBuffer(count); |
There was a problem hiding this comment.
Why? git_buf knows how to grow itself (that is much of the point for it existing).
There was a problem hiding this comment.
Allocating a page would be better yes (assuming you do mean 4kB). The difference in allocations is surprising, since you're also growing by 1.5 and the size is ensured anyhow. The one difference seems to be that you assume there is going to be a next request and preallocate for it. How large is the difference in reallocs? What's the difference in memory consumption?
Warning: This relies on a libgit2 private build from libgit2/libgit2#2892 which isn't merged yet.
Besides the minor fixes, this tries to provide a more stream-like approach. Unfortunately, something's not working and I can't find why.
Ideas?