Message-ID: <gh-postgresql-interfaces-psqlodbc-179@github.com>
From: "jarvis24young (@jarvis24young)" <noreply+jarvis24young@github.com>
To: "postgresql-interfaces/psqlodbc" <noreply+postgresql-interfaces-psqlodbc@github.com>
Date: Tue, 28 Apr 2026 07:11:53 +0000
Subject: [postgresql-interfaces/psqlodbc] PR #179: Handle ARD bookmark allocation failure
List-Id: <gh-postgresql-interfaces-psqlodbc.github.com>
X-GitHub-Author-Id: 48787405
X-GitHub-Author-Login: jarvis24young
X-GitHub-Issue: 179
X-GitHub-Repo: postgresql-interfaces/psqlodbc
X-GitHub-State: merged
X-GitHub-Type: pull_request
X-GitHub-Url: https://github.com/postgresql-interfaces/psqlodbc/pull/179
Content-Type: text/plain; charset=utf-8

## Summary

This hardens ARD bookmark allocation so low-memory failures are reported as ODBC errors instead of dereferencing a NULL pointer.

`ARD_AllocBookmark()` previously called `malloc(sizeof(BindInfoClass))` and immediately cleared the returned pointer. If the allocation failed, statement allocation or bookmark binding could crash the host process.

The patch:

- returns `NULL` from `ARD_AllocBookmark()` when bookmark allocation fails;
- handles that failure in `PGAPI_AllocStmt()` by removing the partially registered statement, destroying it, clearing the output handle, and returning `SQL_ERROR` with the existing statement allocation diagnostic;
- handles the same failure in `SQLBindCol()` bookmark-column binding with `STMT_NO_MEMORY_ERROR`;
- handles ARD bookmark descriptor field updates with `DESC_NO_MEMORY_ERROR`;
- adds a black-box ODBC regression test for the normal statement-allocation path that initializes the ARD bookmark.

## Validation

Built and tested in WSL against the real driver through unixODBC.

Default regression checks:

```sh
cd ~/psqlodbc-current-oom/test
ODBCSYSINI=. ODBCINSTINI=./odbcinst.ini ODBCINI=./odbc.ini \
  ./runsuite ard-bookmark-oom bookmark bulkoperations --inputdir=.
```

Result:

```text
ok 1 - ard-bookmark-oom
ok 2 - bookmark
ok 3 - bulkoperations
```

Manual OOM fault-injection validation with the real driver:

```sh
PODBC_FAIL_MALLOC_SIZE=40 PODBC_FAIL_MALLOC_N=178 \
LD_PRELOAD=./exe/malloc_fail_shim.so \
ODBCSYSINI=. ODBCINSTINI=./odbcinst.ini ODBCINI=./odbc.ini \
  ./exe/ard-bookmark-oom-bugwatch-test
```

Before the fix this produced `Segmentation fault (core dumped)` at `descriptor.c:345` via `statement.c:232`.

After the fix it returns a normal ODBC allocation error without crashing:

```text
SQLAllocHandle stmt rc=-1 hstmt=(nil)
HY001=No more memory to allocate a further SQL-statement
statement allocation returned an error without crashing
```

The LD_PRELOAD fault-injection harness is not registered in the portable regression suite; the committed regression test covers the public ODBC statement-allocation path under normal conditions.