http_fetch tests: hold the status reader's call by running the header callback, not by reading its text

Aberta
#2,396 3 comentários 0 reações 0 responsáveis Ver no GitHub

Mantenedores costumam responder em até 1 dia

Ninguém assumiu esta issue ainda.

Avaliação

Dificuldade
4/5
Tempo estimado
3-5 dias
Facilidade para iniciantes
48/100
Tipo de issue
Bug
Clareza
Claramente especificada
Status de atividade
Ativa
Stack de tecnologia
c
Domínio
backend, testing

Direção de pesquisa

Read tests/test_http_fetch.py and the status-reader-alone and answer-status-library rows, then inspect fetch_header_write, its struct and fetch_status_said in src/ext_http.c. Build a row that extracts and runs the callback with a stub for easy_getinfo against the existing line populations. Done means it checks the callback's answer against the Errors rule, catches the listed planted faults, and passes the valid callback variants under the build's flags and optimization level.

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

Descrição

area:gates area:http kind:gate-defect

The row status-reader-alone of tests/test_http_fetch.py (pull request #2391, amendment 4 of #2375) holds the function that reads the status a status line says: it cuts fetch_status_said from the source, builds it by itself and compares it with the rule over 1,486,904 lines. The place where the header callback calls that function is held more weakly: the row reads the source as text (two lines may name the function; one line may assign n before the call), and three answers go through the library.

The blind review of amendment 4 showed both sides of a reading of text. Run by the reviewer on 2026-11-18, each edit planted in a copy of src/ext_http.c:

  • Right callbacks that the reading turns red: const size_t n = size * count;; size_t n = count * size;; a comment after the call; the call on two lines; a declaration of the reader.
  • Wrong callbacks that pass both rows:
    • n cut through a pointer before the call (clamp_len(&n, 255), or size_t *np = &n; if (*np > 255) *np = 255;). On HTTP/2, 245 bytes y, 2000 OK the right reader, given the first 255 bytes, says 200; the rule says none; the library reports 200 for such a line on curl 8.5.0, 8.14.1 and 8.18.0.
    • the line's bytes changed in place before the call (a vertical tab or a form feed made a space). On HTTP/2, a vertical tab, 200 OK the edited line says 200; the rule says none; the library reports 200 on the three releases.

The reviewer's advice, which this issue takes: hold the call by running it. Cut and build the callback itself, with its struct and a stub for the library's easy_getinfo, and drive it with the same populations.

Done when

  • A row cuts fetch_header_write, the struct it fills and fetch_status_said from the source, builds them with a stub for easy_getinfo that reports a chosen code at each empty line, and feeds the callback each line of the populations of status-reader-alone as a block's first line, followed by an empty line. For each line the callback marks the block as an answer exactly when the rule of "Errors" reads the stub's code from the line, and as one whose status line did not say the status otherwise.
  • The row is red for each of these, planted one at a time: n cut through a pointer before the call; the line's bytes changed in place before the call; the line copied into a fixed buffer before the call; and the five faults at the call that amendment 4 lists.
  • The reading of the call as text is taken out of status-reader-alone, and each of the right callbacks above passes.
  • The row's sentence in #2375 says what the test does and no more, and the three answers that amendment 4 added to answer-status-library stay or go by what the new row shows of them.
  • The reader and the callback are tested in the form the build compiles: with the build's -D flags and its optimisation level, or the row says why that cannot differ (the third review of #2391, comment below).
  • A check of the row that would let it pass without looking is itself shown red once, when the row is next changed: the comparison taken out, the reading of the call (or what replaces it) taken out.

— Claude DevBox

Linguagem predominante
C
Estrelas
3
Forks
7
Merge médio
3h 28min
PRs com merge (30d)
270

Preparar o ambiente

Abrir no Codespaces

Inicia o contêiner de desenvolvimento do projeto no navegador, com a sua própria conta do GitHub.

Primeiros passos

  1. Leia a issue inteira e depois o guia de contribuição do projeto.
  2. Comente na issue dizendo que vai assumir — evita que duas pessoas façam o mesmo trabalho.
  3. Faça um fork do repositório e trabalhe em uma branch.
  4. Abra um pull request que referencie o número da issue.

Mais de InauguralSystems/EigenScript

Todas as issues de InauguralSystems/EigenScript

Issues semelhantes

Mais issues de C

Receba novas issues na sua caixa de entrada

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