Related: #108256

Changelog category (leave one):

• Bug Fix (user-visible misbehavior in an official stable release)

Changelog entry (a user-readable short description of the changes that goes into CHANGELOG.md):

Fix a LOGICAL_ERROR (Column ... already added for reading) when running a query with make_distributed_plan = 1 over a MergeTree table that does a direct read from a text index (hasToken, hasAnyTokens, hasAllTokens). Such reads are now rejected at planning time instead of aborting the server.

Description

Found by the AST fuzzer: AST fuzzer (amd_debug, targeted), STID 4792-6322.
CI report: https://s3.amazonaws.com/clickhouse-test-reports/json.html?PR=108806&sha=70025015949d86cc3d68f7f9f0440ae1a0b6bf18&name_0=PR&name_1=AST%20fuzzer%20%28amd_debug%2C%20targeted%29

Logical error: 'Column __text_index_idx_hasToken_2a3fd3c5b6be32586f14537fc9a3946c already added for reading'

at src/Processors/QueryPlan/ReadFromMergeTree.cpp:4554, on the
buildQueryPipeline -> convertToDistributed -> optimizeTreeSecondPass path.

A direct read from a text index adds ephemeral virtual columns
(__text_index_idx_...) to the read and a separate index-read step. A
distributed worker fragment cannot reproduce these: the column is not in the
table and ReadFromDistributedPlanSource cannot materialize it.

buildQueryPipeline runs the second optimization pass once via optimize()
and, for make_distributed_plan, again via convertToDistributed(). The text
index optimization is not idempotent (it appends the virtual column to
all_column_names), so the single-stage re-optimization re-adds the same
column and aborts with the LOGICAL_ERROR above (a server abort in
debug/sanitizer builds). The multi-stage path instead fails at execution with
NOT_FOUND_COLUMN_IN_BLOCK.

The fix rejects the unsupported combination cleanly at planning time in
checkDistributedReadSupported, next to the existing rejections for pinned
block-number boundaries and the part-order virtual columns. This mirrors the
recently merged #108256, which rejects projections for distributed reads for
the same reason. Reproducer:

SET allow_experimental_full_text_index = 1, make_distributed_plan = 1;
CREATE TABLE t (k UInt64, s String, INDEX idx s TYPE text(tokenizer = 'splitByNonAlpha'))
ENGINE = MergeTree ORDER BY k;
INSERT INTO t SELECT number, 'uniform' FROM numbers(258);
SELECT k FROM t PREWHERE hasToken(s, 'uniform') WHERE hasToken(s, 'uniform');