Two branches, one equals sign

For a moment the repository has two memories of the same afternoon. One of them says hello to Windows. The other adds a JavaScript file whose condition does not ask whether a value is "blah". It makes the value "blah". Philosophy has spent centuries on the difference between naming a thing and changing it. This history spends two days, and then merges the argument.

Scope

This dig examines five commits in bamr87/it-journey from 2020-08-23 through 2020-08-25. They are not one straight line. js and stuff and the already published Hello win share a parent. Two merges follow. This page does not collect pull-request reviews, issue comments, or a test run. The evidence index is docs/git-archaeology/002-two-branches/evidence.json in this repository.

Facts

js and stuff adds js_testing.js and modifies two Mac setup files. In the new file, the loop condition is itemList1[i] = "blah", one equals sign. The parent is 0ef2f16839b8ff1421ba3980363da97d9859e09f, the same parent as Hello win. The committer is amr abdel-motaleb. git-commit:b6d3cf5958e6c639f13b60f47f62af715f3e788d

Fifty-one seconds later, Merge branch 'master' of https://github.com/Bamrman/it-journey joins Hello win and js and stuff. The author and committer are Bamrman. The merge commit itself has an empty name-status. git-commit:e5f9d75ed0130be054310daba72a03390fa00b78

update does not sit on that merge. Its parent is js and stuff. It changes the condition to itemList1[i] == "blah" and maps the quantity strings with .map(. It also rewrites the init_it.md headings from Level 00 through Level 02 into binary labels, Level 0000 through Level 1111. The last of those headings is Level 1111 - Borne Again Solutions Hero. The heading Level 0010 - PIjects keeps the earlier spelling. git-commit:1cae4d51b37a7a875855dd6acdd6ca5503562e5e

comit sits on the other side, parented by the first merge. It still has itemList1[i] = "blah", and it changes the addition to valueOf. It adds an empty gpc-app.md and edits hello-win.md. The author and committer are Bamrman. git-commit:244e56725cef6fa4dec242367e882244a700e45d

Merge branch 'master' of github.com:Bamrman/it-journey joins update and comit. Its only recorded path is js_testing.js, with status MM. The file stored at that commit contains itemList1[i] = "blah" and does not contain the two-equals comparison. It also contains no conflict markers. The committer is amr abdel-motaleb. git-commit:4304ba962ea09e180c0249eed686587cd547d2a3

Inferences

The two equals-sign versions are not a before-and-after on one line. They are contemporaries. One branch compares. The other still assigns. The merge tip keeps the assigning form. That is what the stored file shows. It does not show whether a person chose that line or a merge tool did.

The binary level headings appear in the same commit as the comparison fix. That pairing suggests the outline was being renamed toward the later 0000 to 1111 ladder. The commit does not say the ladder was taught, published, or followed. PIjects surviving the rename suggests the edit was a relabel, not a proofread. The correction and the ladder both lose the argument at the tip: the comparison is not the version this sample ends on.

Limits

This sample cannot say the JavaScript was executed, or why the assigning form is the one at the merge tip. It cannot say the empty gpc-app.md was a typo for GCP, only that the filename is gpc-app.md and the file is empty. The open question is what a history owes a reader when the careful correction and the surviving bug are both true, and only one of them remains.

Retrospective

This is hindsight, written now. The merge does not say anyone noticed the loss.

I wish the condition had been a comparison on the first commit, and I wish a check had refused an assignment inside if. The bug is one character. The time cost is the rest of this dig. git-commit:b6d3cf5958e6c639f13b60f47f62af715f3e788d

I wish the comparing branch had been the one merged, or the two lines had been rebased onto one parent before either side called itself done. Instead the tip stores the assigning form after a commit had already written ==. The lesson is not “merges are bad.” It is that a correction on a side branch is not a correction until the tip contains it. git-commit:1cae4d51b37a7a875855dd6acdd6ca5503562e5e git-commit:4304ba962ea09e180c0249eed686587cd547d2a3

I wish the empty gpc-app.md had not been added under a misspelled name. An empty file teaches nothing and leaves the next reader to guess. Not adding it would have saved that guess. git-commit:244e56725cef6fa4dec242367e882244a700e45d