allow adding header and footer, autoconfirm, add js-tiktoken, add faster check for going to file-only mode #343
Conversation
refactor: remove unused fs operations from esbuild.config.js chore: update dependencies and add overrides in package.json docs: update commit message guidelines in prompts.ts refactor: switch to js-tiktoken for token encoding in tokenCount.ts
chore(config): add new config keys for prompt header and footer style: reformat code for better readability and consistency across files refactor(prompts.ts): restructure prompt generation for clarity and maintainability chore: update dependencies and improve environment variable handling style(server.ts): rename port variable to uppercase for consistency and clarity chore(server.ts): add support for configurable port via process.env.PORT style(engine.ts, tokenCount.ts, version.ts): standardize import quotes to single quotes for consistency
… directory chore(package.json): add tiktoken dependency style(config.ts, prompts.ts): fix formatting and indentation issues refactor(tokenCount.ts): add native token counting using Tiktoken and conditional export based on environment variable
chore(commit.ts): add check for empty remotes array before selecting remote perf(tokenCount.ts): move encoding initialization inside function to reduce memory usage
…ore than one remote is available
Introduce OCO_GIT_STAGE_ALWAYS config key to bypass user confirmation for staging all files. This enhances automation and streamlines the commit process when the config is set to true.
…ild configuration
refactor(commit.ts): add auto-confirmation for commit and stage actions refactor(config.ts): rename OCO_GIT_STAGE_ALWAYS to OCO_AUTOCONFIRM_STAGE and add OCO_AUTOCONFIRM_COMMIT fix(openAi.ts): handle undefined message content in token count style(prompts.ts): move OCO_PROMPT_FOOTER to the end of the prompt style(mergeDiffs.ts): adjust import statement and spacing in token count check
…ove error handling
|
if you want this to be merged at some point, please describe the changes, this will help me to review the code |
|
i have some more changes and cleanup i would want to make, as i added the features just to make them work for me, and i wouldn't want to merge half baked hacks into your repo. if i find some time to make this merge ready, i will probably redo much of it, because i am not happy with the hacky solutions i came up with. regardless, i'll still detail what i did in case it might be useful for your own development FWIW, you can fix the punycode warning by adding this version override in package.json otherwise, these changes two solve two issues with this app for my usage:
i monkey patch this by adding a function which estimates the token count based off character count, but i think this is a bad solution and probably shouldnt be merged as it currently stands. it is good enough for my use cases. for instance, if the diff is 20,000 characters, it most likely will not fit in a 1000 token context.
my hack was to add PREFIX/SUFFIX env vars to patch the prompt for the behavior i wanted, but i feel this is a bad solution, and really the entire prompt builder and configuration around prompt generation needs to be redone for me to feel happy about merging. im happy with my hack for personal use, but i really dont think it is something that people should use. the way that the prompt strings are created in the existing library is a mess. the repeated code across the two different prompt types really bothers me, along with the lack of customization around the prompt in general. |
lots of changes in this pr that ive been making for myself as ive been using the tool.
dont really think its ready to merge but thought i would put it here
it also fixes the punycode warning.