I've a spec which has
table: data/csv/a.csv
Tags: tag1
Where csv file has 49 cases and there is only 1 scenario in spec
Before spec in a StemImplementation looks like below
@BeforeSpec(tags = {"tag1","tag2"},tagAggregation = Operator.OR)
public static void preSpec() {
}
When i run spec in parallel, it is running before spec 49 times. It should be run as many thread i've.
gauge -v
Gauge version: 1.1.1 Plugins
html-report (4.0.12)
java (0.7.3)
json-report (0.3.3)
xml-report (0.2.3)
There is an experimental feature on gauge - Table driven scenario. Where table can be defined at scenario level and then beforespec and afterspec will run only once. Complete spec will run in one go.
In parallel mode, all specs will run in parallel instead of scenario. This is what people mostly want
More details here https://docs.gauge.org/writing-specifications.html?os=linux&language=javascript&ide=vscode#table-driven-scenario
There are open issues with this approach.