…rch-results-java

- Fix SerpApi homepage URL to use proper markdown link syntax
- Add migration guide section covering dependency change, class/method
  renames, and before/after code example for users upgrading from the
  legacy google-search-results-java library

Co-Authored-By: Claude Sonnet 4.6 <[email protected]>

The google_events engine returned no events_results for q=coffee, so
getAsJsonArray returned null and .size() threw an NPE, failing CI.

Use a location-scoped query, assert non-null with the response body in
the message so future failures are diagnosable, and relax the size
threshold from >5 to >0.

Co-Authored-By: Claude Opus 5 <[email protected]>

exceptionFormat = "short" prints only the exception class and line,
swallowing assertion messages. GoogleEventsTest embeds the API response
in its assertion message, which CI was discarding, making the failure
undiagnosable from logs alone.

Co-Authored-By: Claude Opus 5 <[email protected]>

The google_events engine responds HTTP 200 / status "Success" but with
events_results_state "Fully empty" and an error field saying Google
returned no results. Reproduced across two queries two days apart, so
it is the engine rather than the query or this client.

Ignore the test so CI reflects the state of this library rather than an
upstream outage. Re-enable once the engine returns events again.

Co-Authored-By: Claude Opus 5 <[email protected]>

SerpApi reports some failures inside the body of an HTTP 200 response,
for example the google_events engine:

  HTTP 200
  {"search_metadata":{"status":"Success"},
   "search_information":{"events_results_state":"Fully empty"},
   "error":"Google hasn't returned any results for this query."}

The client decided success purely from the status code, so search()
returned an object that was semantically an error. Callers then reached
for the key they expected and got a NullPointerException pointing at
their own code, with the explanation sitting unread in the error field.
This is what broke GoogleEventsTest.

Check for a body-level error in json() and location(), routing it
through the existing triggerSerpApiException so every SerpApi error
reaches the caller as a SerpApiException regardless of status code.

html() still returns its raw String unchecked; parsing arbitrary HTML
as JSON to look for an error field is not worth the risk.

Co-Authored-By: Claude Opus 5 <[email protected]>

SerpApi accepts output=md, "a markdown-formatted version optimized for
LLMs and AI agents", but this client exposed no way to ask for it short
of calling the low-level get("/search", "md", parameter) and knowing the
endpoint path.

Add markdown(), mirroring html(): the raw String is returned unparsed.
Also correct the get() javadoc, which listed the output formats as
(json, html, json_with_images) where the API documents (json, html, md).

Tests stub the HTTP client and assert on the request this client builds,
so they run offline without a SERPAPI_KEY.

Co-Authored-By: Claude Opus 5 <[email protected]>

Drop the intro and trailing note, and follow the auth/parameter naming
and the "it prints ..." closing line used by the surrounding sections.

Co-Authored-By: Claude Opus 5 <[email protected]>

Bump the project version and the places that restate it:

- build.gradle, which names the published artifact
- README installation snippet and changelog, in both the erb template
  and the generated README.md
- SerpApiHttp.VERSION, which had drifted to 1.0.0 while the project was
  at 1.1.0. It is public and never read anywhere in the library, so the
  stale value was only ever visible to callers.

demo/build.gradle still pins 1.0.0 and is left alone, since it resolves
a published JitPack artifact rather than this source tree.

Co-Authored-By: Claude Opus 5 <[email protected]>

Co-Authored-By: Claude Opus 5 <[email protected]>

HomeDepotTest failed 4 of the last 7 CI runs, always the same way:

  SerpApiException: java.net.http.HttpTimeoutException: request timed out

SerpApi.timeout was wired only to setHttpConnectionTimeout, so the read
timeout was never configured and kept SerpApiHttp's 60s default. That is
backwards: a 60s connection timeout is meaningless, since connecting
takes milliseconds, while 60s to read is tight for engines that scrape.
Apply the timeout to both, and raise the default to 120s, which the
home_depot engine regularly needs.

This is a library fix rather than a test fix: any caller searching a slow
engine hit the same timeout.

Also repair the README snippet helper, which sliced examples at a
hardcoded lines[23..], assuming every example body starts at line 24.
Quarantining GoogleEventsTest shifted that file and pulled the @ignore
and its comment into the docs as if they were usage code. Anchor on the
"// setup serpapi client" marker instead, which every example test has
exactly once, and regenerate.

Co-Authored-By: Claude Opus 5 <[email protected]>

Created accidentally by a shell redirect run from the wrong directory.

Co-Authored-By: Claude Opus 5 <[email protected]>