HTTP tests: since the new request reader, 50 of the request-header row's 390 dealt requests reach a route

Aberta
#2,431 1 comentário 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
1-2 dias
Facilidade para iniciantes
56/100
Tipo de issue
Bug
Clareza
Claramente especificada
Status de atividade
Ativa
Stack de tecnologia
c
Domínio
testing

Direção de pesquisa

Start with tests/test_http_request_header.py and inspect the QH01-dealt row, its seeded deal, and the request helpers; compare request outcomes using the reference in tests/test_http_door.py. Adjust the deal and assertions so each reachable header kind has at least 15 requests that reach a route, while unreachable kinds have explicit refusal counts or are excluded. Done means the row reports served counts accurately, unreachable reading rules are removed or tested directly, and the described planted faults make the row fail.

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

Descrição

area:http kind:gate-defect

Found while checking #2373 box by box on phase2-numbers at c96498d9.

The row QH01-dealt of tests/test_http_request_header.py deals 390 requests, 15 of each of 26 kinds of header line, and compares each answer byte for byte with a reference. It was written before the server read every request through its new request reader (#2370). Each dealt request holds one line of its kind and up to ten more lines of random kinds, and the reader now refuses a whole request with 400 when any one line is malformed. So most of the population is now a test of that refusal, which the reader's own test already covers, and little of it reaches http_request_header:

  • 50 of the 390 requests reach a route; 340 are answered 400 before any route runs.
  • 11 of the 26 kinds never reach a route (kinds 0, 11, 12, 13 and 19 to 25).
  • The other 15 kinds reach a route 2 to 5 times each, not 15.

The row still passes and still prints dealt requests=390/390, and its count check ("390, and every kind 15") is about what was dealt, not about what was served. A fault in how http_request_header reads a well-formed line has far fewer requests to show itself on than the row's numbers say.

Reproduction, from tests/ of a checkout of that commit: replay the row's seeded deal through the reference of tests/test_http_door.py and count what it decides.

import random
from collections import Counter
import test_http_request_header as t
from test_http_door import shared_reference
ref = shared_reference(); rng = random.Random(t.SEED)
kinds = list(range(26)) * 15; rng.shuffle(kinds)
posts = set(rng.sample(range(390), 195))
bodies = (b'', b't' * 40, b'x\r\nOrigin: https://b.example\r\n', b'b' * 20 + b'\r\n\r\nCookie: a=b\r\n')
served = Counter()
for i, kind in enumerate(kinds):
    others = [t.dealt_line(rng, rng.randrange(26)) for _ in range(rng.randrange(11))]
    others.insert(rng.randrange(len(others) + 1), t.dealt_line(rng, kind))
    post = i in posts
    data = t.request(others, rng.choice(bodies) if post else b'', method=b'POST' if post else b'GET')
    if ref.decide(data).kind == 'served': served[kind] += 1
print(sum(served.values()), [served[k] for k in range(26)])

It prints 50 [0, 3, 4, 3, 2, 4, 3, 4, 2, 2, 4, 0, 0, 0, 3, 3, 4, 4, 5, 0, 0, 0, 0, 0, 0, 0].

Done when

  • The row asserts a count of requests that reached a route, for each kind that can reach one, and that count is at least 15. The deal is changed so that it holds (for example: the other lines of a request come only from kinds the reader accepts).
  • Each kind that can no longer reach a route is either dealt in a part of the row that expects the refusal, with its own count, or taken out of the deal. The row's printed counts say which.
  • For each reading rule of http_request_header that no request can reach any more (a control byte or a lone CR or LF in a value, a line with no colon, a bare name, a space before the colon, a continuation line, an empty name): it is removed, or it is kept and a test calls the reading code with such a line directly. The choice is written here.
  • A planted fault in how a well-formed line is read (one for the name's case, one for the value's trimming) turns the row red, shown once in the pull request.

— 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.