From 0adef6e16a9c1d65131c97b3c8f0bc19aef7d28c Mon Sep 17 00:00:00 2001 From: zipg Date: Sat, 19 Sep 2026 15:23:22 +0800 Subject: [PATCH] =?UTF-8?q?fix(export):=20=E5=88=86=E9=A1=B5=E5=AF=BC?= =?UTF-8?q?=E5=87=BA=E6=8C=89=E5=88=97=E5=90=8D=E5=AF=B9=E9=BD=90=E6=AF=8F?= =?UTF-8?q?=E4=B8=80=E9=A1=B5=EF=BC=8C=E4=BF=AE=E5=A4=8D=E5=AD=97=E6=AE=B5?= =?UTF-8?q?=E9=94=99=E4=BD=8D=20(#9583)?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Schemaless 引擎(Elasticsearch/Easysearch)的列集合是按每个响应里的文档现算的, 因此 select * 的不同分页可能给出不同的列顺序或列集合。查询结果导出把表头冻结在 第一页,却按位置写入后续页的行,导致第二页起的值落到错误的列里(#9522 反馈的 "导出字段错位",例如 beta 的值被写进 alpha 列)。 - 新增 export_column_remap/realign_page_cells/realign_page_rows:后续页按列名映射到 冻结表头,列名缺失时填 NULL,列顺序不同也能正确还原; - 列类型同样按列名对齐,保证按页的日期/时间格式化使用正确的类型; - spatial_values 随行一起对齐,保持与单元格的 SRID 对应关系; - 第一页仍是表头来源,新增列(表头里没有的列)与被删除的列都会被忽略,行为与写出 表头保持一致。 Co-authored-by: zipg <4047349+zipg@users.noreply.github.com> --- .../dbx-core/src/data/query_result_export.rs | 73 ++++++++++++++++++- 1 file changed, 70 insertions(+), 3 deletions(-) diff --git a/crates/dbx-core/src/data/query_result_export.rs b/crates/dbx-core/src/data/query_result_export.rs index 201bc5e9c7..fe08ec3096 100644 --- a/crates/dbx-core/src/data/query_result_export.rs +++ b/crates/dbx-core/src/data/query_result_export.rs @@ -630,6 +630,27 @@ fn single_execution_row_bound(request: &QueryResultExportRequest) -> Option Vec> { + header_columns.iter().map(|column| page_columns.iter().position(|page_column| page_column == column)).collect() +} + +/// Reorders one page row (or a page-level cell list such as column types) into +/// the frozen header order (`remap` indexes the page), filling cells whose column +/// the page did not report with the cell type's default (`NULL` for values, empty +/// for column types). +fn realign_page_cells(remap: &[Option], cells: Vec) -> Vec { + remap.iter().map(|index| index.and_then(|index| cells.get(index)).cloned().unwrap_or_default()).collect() +} + +/// Applies [`realign_page_cells`] to every row of a page. +fn realign_page_rows(remap: &[Option], rows: Vec>) -> Vec> { + rows.into_iter().map(|row| realign_page_cells(remap, row)).collect() +} + fn single_execution_page_limit(request: &QueryResultExportRequest, page_size: usize) -> Option { single_execution_row_bound(request).filter(|bound| *bound > 0 && *bound <= page_size.max(1)) } @@ -973,13 +994,27 @@ async fn export_query_result_core_inner( return Ok(()); } - if columns.is_empty() { + let page_column_types = if columns.is_empty() { columns = result.columns.clone(); column_types = result.column_types.clone(); if let Some(writer) = sql_writer.as_mut() { writer.set_columns(columns.clone(), &column_types, &result.spatial_columns, request); } - } + column_types.clone() + } else if result.columns == columns { + column_types.clone() + } else { + // The export header is frozen by the first page, so every later page + // must be aligned by column name; writing page rows positionally + // dropped/duplicated values into the wrong columns. + let remap = export_column_remap(&result.columns, &columns); + let page_column_types = realign_page_cells(&remap, result.column_types.clone()); + result.rows = realign_page_rows(&remap, mem::take(&mut result.rows)); + if !result.spatial_values.is_empty() { + result.spatial_values = realign_page_rows(&remap, mem::take(&mut result.spatial_values)); + } + page_column_types + }; let fetched_row_count = result.rows.len(); if result.rows.len() > this_page { result.rows.truncate(this_page); @@ -987,7 +1022,7 @@ async fn export_query_result_core_inner( let row_count = result.rows.len(); let formatted_rows = crate::temporal_format::format_temporal_export_rows_with_string_types_cow( &result.rows, - &column_types, + &page_column_types, request.date_time_format.as_deref(), ); @@ -1980,6 +2015,38 @@ mod tests { use super::*; use serde_json::json; + #[test] + fn export_column_remap_follows_column_names_across_pages() { + // Elasticsearch-style page drift: the second page reports `beta` where the + // header has `alpha`, so the `alpha` cell must be filled with NULL instead + // of the value that happens to sit at the same position. + let header = vec!["id".to_string(), "alpha".to_string(), "_id".to_string()]; + let page = vec!["id".to_string(), "beta".to_string(), "_id".to_string()]; + let remap = export_column_remap(&page, &header); + assert_eq!(remap, vec![Some(0), None, Some(2)]); + let rows = vec![vec![json!(101), json!("B101"), json!("101")]]; + assert_eq!(realign_page_rows(&remap, rows), vec![vec![json!(101), Value::Null, json!("101")]]); + } + + #[test] + fn export_column_remap_reorders_pages_that_report_columns_differently() { + let header = vec!["id".to_string(), "alpha".to_string()]; + let page = vec!["alpha".to_string(), "id".to_string()]; + let remap = export_column_remap(&page, &header); + assert_eq!(remap, vec![Some(1), Some(0)]); + let rows = vec![vec![json!("A1"), json!(1)]]; + assert_eq!(realign_page_rows(&remap, rows), vec![vec![json!(1), json!("A1")]]); + } + + #[test] + fn export_column_remap_keeps_identical_headers_untouched() { + let header = vec!["id".to_string(), "alpha".to_string()]; + let remap = export_column_remap(&header, &header); + assert_eq!(remap, vec![Some(0), Some(1)]); + let rows = vec![vec![json!(1), json!("A1")]]; + assert_eq!(realign_page_rows(&remap, rows.clone()), rows); + } + #[test] fn staged_export_target_preserves_existing_destination_on_discard_and_replace_failure() { let dir = tempfile::tempdir().expect("temp dir");