`Include_all_commits` sometimes shows incorrect number of commits

[nxwi]

Describe the bug

On GitHub Stats the number of commits is currently showing the last years commits, after adding include_all_commits it decreases as shown below.

Expected behaviour

Shows the number of commits in 2022

Screenshots / Live demo link (paste the github-readme-stats link as markdown image)

Before adding include_all_commits

After adding include_all_commits

[rickstaa]

@nasw1j, I checked the codebase, and I think both count_private and include_all_commits parameters are correctly implemented.

github-readme-stats/src/fetchers/stats-fetcher.js

Lines 155 to 163 in 9a909ff

	if (include_all_commits) {
	stats.totalCommits = await totalCommitsFetcher(username);
	}
	// if count_private then add private commits to totalCommits so far.
	if (count_private) {
	stats.totalCommits +=
	user.contributionsCollection.restrictedContributionsCount;
	}

I, therefore, think the error is caused by incorrect data returned from GitHub's REST API and GraphQL API APIs. If this happens, these wrong values are cached for 4 hours.

github-readme-stats/api/index.js

Lines 60 to 64 in 9a909ff

	const cacheSeconds = clampValue(
	parseInt(cache_seconds || CONSTANTS.FOUR_HOURS, 10),
	CONSTANTS.FOUR_HOURS,
	CONSTANTS.ONE_DAY,
	);

When this happens, you can check the console.log to see if there is any error and use the &v=123 syntax to force the cache to reload.

Detailed explanation

If include_all_commits is enabled, and GitHub REST API fails the code will return 0.

github-readme-stats/src/fetchers/stats-fetcher.js

Lines 69 to 100 in 9a909ff

	const totalCommitsFetcher = async (username) => {
	if (!githubUsernameRegex.test(username)) {
	logger.log("Invalid username");
	return 0;
	}
	// https://developer.github.com/v3/search/#search-commits
	const fetchTotalCommits = (variables, token) => {
	return axios({
	method: "get",
	url: `https://api.github.com/search/commits?q=author:${variables.login}`,
	headers: {
	"Content-Type": "application/json",
	Accept: "application/vnd.github.cloak-preview",
	Authorization: `token ${token}`,
	},
	});
	};
	try {
	let res = await retryer(fetchTotalCommits, { login: username });
	let total_count = res.data.total_count;
	if (!!total_count && !isNaN(total_count)) {
	return res.data.total_count;
	}
	} catch (err) {
	logger.log(err);
	}
	// just return 0 if there is something wrong so that
	// we don't break the whole app
	return 0;
	};

This will then be used as the total commits. To this value, the private commits will be correctly added if count_private is enabled. Since we used caching, this wrong value will be on the card for 4 hours.

Posible solutions

We can use the previous cache when this problem happens.

[rickstaa]

I quickly checked #1691 created by @Rongronggg9 and think merging to his pull request will significantly improve the commit accuracy. I, however, did not have time to do a thorough review 😅.

Nevertheless, still, one problem persists with #1691 merged. It will show a commit count of 0 when the GitHub Rest API fails. @Rongronggg9 maybe, if the Rest API fails, we can use the old cache or prevent the wrong result from being cached?

github-readme-stats/api/index.js

Line 66 in 9a909ff

res.setHeader("Cache-Control", `public, max-age=${cacheSeconds}`);

Maybe this can be achieved by changing the cache header when the Rest API fails (see https://vercel.com/docs/concepts/edge-network/caching).