config read once

[tusharmath]

• chore(ISSUE_TEMPLATE): streamline bug report fields
• feat(ui): add workspace init confirmation before syncing (feat(ui): add workspace init confirmation before syncing #2822)
• feat(provider): add Google AI Studio support (chore(provider): add Google AI Studio support #2805)
• feat(auth): add OAuth authorization code + PKCE for Codex provider (feat(auth): add OAuth authorization code + PKCE for Codex provider #2790)
• chore(deps): bump the major group across 1 directory with 5 updates (chore(deps): bump the major group across 1 directory with 5 updates #2783)
• fix(bedrock): sanitize tool call IDs for Anthropic compatibility (fix(bedrock): sanitize tool call IDs for Anthropic compatibility #2837)
• feat(config): read ForgeConfig once at startup with cache

[tusharmath]

Completely drop ConfigInfra. We will never read config infra dynamically ever. Make sure to pass an instance of ForgeConfig in <TypeName>::new() constructor wherever required.

[tusharmath]

Revert all changes in forge_config.

[tusharmath]

Keep context, but don't provide file path. It can change we don't want to hardcode it here. Just say .forge.toml

[graphite-app]

Critical: .expect() will panic if config re-read fails after update_environment

The get_config() implementation uses .expect() on a Result<ForgeConfig> that can fail. After update_environment() invalidates the cache (line 165), the next get_config() call triggers a disk re-read via cached_config(). If that re-read fails (e.g., corrupted TOML file, permissions issue), the application will panic instead of returning an error.

Fix: Change the trait signature to return Result<ForgeConfig> and propagate the error, or handle the error gracefully:

impl ConfigReaderInfra for ForgeEnvironmentInfra {
    fn get_config(&self) -> ForgeConfig {
        self.cached_config()
            .unwrap_or_else(|e| {
                tracing::error!(error = ?e, "Failed to read config after update; using defaults");
                ForgeConfig::default()
            })
    }
}

Alternatively, make the trait method fallible:

fn get_config(&self) -> anyhow::Result<ForgeConfig>

Suggested change

	impl ConfigReaderInfra for ForgeEnvironmentInfra {
	fn get_config(&self) -> ForgeConfig {
	self.cached_config()
	.expect("ForgeConfig cache read failed after update_environment")
	}
	}
	impl ConfigReaderInfra for ForgeEnvironmentInfra {
	fn get_config(&self) -> ForgeConfig {
	self.cached_config()
	.unwrap_or_else(|e| {
	tracing::error!(error = ?e, "Failed to read config after update; using defaults");
	ForgeConfig::default()
	})
	}
	}

Spotted by Graphite



Is this helpful? React 👍 or 👎 to let us know.

[graphite-app]

Defeats the purpose of the PR by silently falling back to defaults on config read errors.

The line reads:

let config = forge_config::ForgeConfig::read().unwrap_or_default();

This silently swallows configuration errors using unwrap_or_default(), which is exactly the behavior this PR is supposed to fix. If the config file is corrupt when the user runs /new, it will silently use default values (all zeros) instead of surfacing the error to the user.

Fix:

let config = forge_config::ForgeConfig::read()
    .context("Failed to read configuration for new conversation")?;

This propagates the error so the user sees a clear message instead of silently broken behavior with default config values.

Suggested change

	let config = forge_config::ForgeConfig::read().unwrap_or_default();
	let config = forge_config::ForgeConfig::read().context("Failed to read configuration for new conversation")?;

Spotted by Graphite



Is this helpful? React 👍 or 👎 to let us know.