We have a large code base and it takes hours to index. Therefore, we index the code base using ccls in cron at night. We are using the btrfs file system which means we can create instant snapshots (clones). So our workflow is
cron at night: run ccls into a "latest" area and store the index within the "lastest" directory
workday: clone the latest area into a "working" area, which means we copied the pre-generated index
Using emacs as a front end, we work in our "working" area. We start LSP / CCLS using the pathMappings option to reuse the copied pre-generated index
We've noticed that if trackDependency is 0 we see in the Emacs *ccls::stderr* thousands of "I load cache for" messages and ccls starts up quickly in a few minutes.
However, if we set trackDependency is 2, we see in the Emacs *ccls::stderr* thousands of "I load cache for" and "I store index for" messages. In this case ccls starts up slowly taking hours. I assume the "I store index for" messages are CCLS re-indexing the file that is already index.
Why does changing the value of trackDependency not reuse the already computed index?
You should be able to reproduce this by
pre-indexing say ccls itself in a "latest" directory
cloning or cp -r the "latest" directory to a "working" directory
Use emacs or other editor to start ccls with the different values of trackDependency and the necessary pathMappings.
System information
ccls version (git describe --tags --long): CCLS from Dec 2, 2022, ec101b92983066b652921ab0702e1b76619aa201
clang version: 15.0.0
OS: Debian 11
Editor: Emacs 27
Language client (and version): Nov-30-2022 version, 2dd33404eb8f7133683b94e18c937de4707d4c33
We have a large code base and it takes hours to index. Therefore, we index the code base using ccls in cron at night. We are using the btrfs file system which means we can create instant snapshots (clones). So our workflow is
We've noticed that if trackDependency is 0 we see in the Emacs
*ccls::stderr*thousands of "I load cache for" messages and ccls starts up quickly in a few minutes.However, if we set trackDependency is 2, we see in the Emacs
*ccls::stderr*thousands of "I load cache for" and "I store index for" messages. In this case ccls starts up slowly taking hours. I assume the "I store index for" messages are CCLS re-indexing the file that is already index.Why does changing the value of trackDependency not reuse the already computed index?
You should be able to reproduce this by
System information
git describe --tags --long): CCLS from Dec 2, 2022, ec101b92983066b652921ab0702e1b76619aa201