apache / apache/datafusion-sqlparser-rs

COPY TO/FROM is overly restrictive on options

Aberta
#1,085 4 comentários 0 reações 0 responsáveis Ver no GitHub
Linguagem predominante
Rust
Estrelas
3.5k
Forks
772
Merge médio
4d 9h
PRs com merge (30d)
17

Descrição

Current parsing of `COPY TO/FROM` options is too restrictive on the how the keys & values can be provided.

https://github.com/sqlparser-rs/sqlparser-rs/blob/ce498864dc705f72e9f85d4dc5f7eba3d17b9ef6/src/parser/mod.rs#L5157-L5180

Given this parsing code, the expected format is:

```sql
COPY (SELECT 1 AS a, 2 AS b) TO 'file'
WITH (
FORMAT 'text',
FREEZE true,
DELIMITER 'd',
NULL 'null',
HEADER true,
QUOTE 'q',
ESCAPE 'e',
FORCE_QUOTE (a, b),
FORCE_NOT_NULL (a, b),
FORCE_NULL (a, b),
ENCODING 'encoding'
)
```

The problems:

- Key must be a keyword and cannot be a quoted string, even though postgres allows this, i.e. `WITH ("format" 'text', ...)`
- It strictly limits the key set at parsing time, leading to further problems:
- Need to update this parser whenever postgres supports a new key, e.g. DEFAULT (supported in postgres 16: https://www.postgresql.org/docs/16/sql-copy.html)
- We can't use this code for other databases/engines with a similar copy statement as they may have different supported keys (e.g. [duckdb](https://duckdb.org/docs/sql/statements/copy.html) and datafusion, see #1080)
- Some values are too strict in their format:
- FORMAT expects an identifier, which disallows escaped strings (`format e'csv'`), but this is allowed by postgres
- FREEZE expects keywords `true` or `false` but postgres allows these to be passed as string literals (single quoted, double quoted, escaped), e.g. `freeze 'true'`
- Same for HEADER

**Solutions**

We could fix these problems for Postgres specifically to make it more permissive, but I was hoping to also fix it to allow usage by other engine dialects (e.g. duckdb and datafusion). This might involve removing the `CopyOption` enum entirely:

https://github.com/sqlparser-rs/sqlparser-rs/blob/ce498864dc705f72e9f85d4dc5f7eba3d17b9ef6/src/ast/mod.rs#L4707-L4736

And replacing its usage in `Statement::Copy` with a more generic version that allows any string key with a new value enum to represent different values (e.g. string value, parenthesized column list value, number value):

https://github.com/sqlparser-rs/sqlparser-rs/blob/ce498864dc705f72e9f85d4dc5f7eba3d17b9ef6/src/ast/mod.rs#L1466-L1478

I'm not sure how the legacy options will fit into this, it could just be left as is and probably tackled as part of another ticket (as the original reason here was for interop with datafusion copy statement)

Guia de contribuição

Nenhum guia de contribuição indexado para este repositório

Direção de pesquisa

Comece pelo parsing das opções de COPY em src/parser/mod.rs nas linhas referenciadas e pelas definições de CopyOption e Statement::Copy em src/ast/mod.rs. Compare a representação atual com os exemplos de opções de PostgreSQL, DuckDB e DataFusion na issue, incluindo chaves entre aspas e valores permissivos. Considera-se concluído quando as opções de COPY forem genéricas o suficiente para esses dialetos sem restringir desnecessariamente as chaves ou os valores, e as opções legadas continuarem sendo contempladas.

Escrita pelo modelo de indexação a partir do texto da issue.

Avaliação

Stack de tecnologia
rust
Domínio
compilers
Tipo de issue
Bug
Dificuldade
5/5
Tempo estimado
Mais de uma semana
Status de atividade
Estagnada
Clareza
Razoavelmente clara
Facilidade para iniciantes
30/100

Receba novas issues na sua caixa de entrada

Um resumo curto de issues do GitHub para quem está começando.