In Allure 2, the previous report's `history/` directory is what gives the next report a trend. In which direction does that copy go, and why must it run before `allure generate` rather than after?

## Two directories, two roles `allure generate` has exactly one input shape and one output shape. It reads one or more **results directories** -- the folders the test run filled with `*-result.json`, `*-container.json` and attachment files -- and it writes a **report directory**, by default `allure-report`, holding a self-contained static site. Nothing else is consulted. In particular, **the generator never reads its own output directory as a source.** That single fact settles both halves of the carry step. ## Direction: out of the report, into the results A generated report contains a `history/` directory of its own, written by the history plugins during generation. It holds the series as data: - `history.json` -- the run-by-run block behind an individual test case's history - `history-trend.json` -- the run-over-run trend series - `duration-trend.json` and `categories-trend.json` -- the other series, written into the same folder That folder is simultaneously **this run's output** and **the next run's input**. To be read, it has to move to the input side: ```bash cp -R ./allure-report/history ./allure-results/history allure generate ./allure-results -o ./allure-report ``` Nothing is renamed, merged or converted on the way. The generator resolves `<results>/history/<name>.json` using exactly the file names it wrote under `history/` in the report, so a recursive directory copy is the entire operation. Leaving the folder in the report directory feels reasonable -- it is already sitting there -- and it accomplishes nothing. The generator will not look at it, and the next generation into that directory rebuilds the contents regardless. ## Timing: before the pass, not after Generation is a single pass: read the results, compute, serialise the report. Each trend series is computed *during* that pass, from the carried history plus this run's results. So a `history/` folder that arrives after the generator has exited is not merely late -- it is not read at all, and the report already on disk will not change. | when the copy runs | what the generator sees | what the report shows | |---|---|---| | before generation, into the results directory | the carried series plus this run | the full trend | | after generation, into the results directory | nothing | a single point | | any time, into the report directory | nothing | a single point | The last two rows are the trap. All three cases exit successfully and produce a valid report; only the chart differs. ## Why neither mistake announces itself The generator's history read is a plain file-existence check with no failure branch. When `<results>/history/history.json` is absent, the plugin contributes nothing and generation continues to a clean exit. That is deliberate -- the genuine first run of a suite legitimately has no history -- but it means a copy in the wrong direction, a copy at the wrong time, and a real first run all emit identical signals: a green build and a correct-looking report. The failure is therefore diagnosed from artefacts, not from exit codes: 1. Confirm `history/history.json` exists in the **results** directory immediately before generation runs. 2. Confirm the generator was pointed at that same results directory and not a sibling of it. 3. Open the produced report's `history/history.json` and count the entries recorded for a long-lived test case. That number is the real depth of the carried series. ## Practical shape of the step Three properties make the carry robust: - **It is a directory copy, not a file copy.** The series files travel together, so carrying `history.json` alone yields per-case history with no trend chart behind it. - **It is self-perpetuating in the right order.** Copy first, generate second, and the report you have just produced is itself the correct source for the next build. - **Its absence should be made loud.** Since the generator will not complain, an explicit check that the folder landed -- with an exemption for a first run -- is the only thing that converts a silently truncated trend into a visible failure. ## Summary The `history/` directory travels **out of the finished report and into the next run's results directory**, and it has to be there **before** `allure generate` starts. Both constraints fall out of the same property: the generator reads results and writes a report, in one pass, and never the other way round.